topbanner_forum
  *

avatar image

Welcome, Guest. Please login or register.
Did you miss your activation email?

Login with username, password and session length
  • Monday December 9, 2024, 11:36 pm
  • Proudly celebrating 15+ years online.
  • Donate now to become a lifetime supporting member of the site and get a non-expiring license key for all of our programs.
  • donate

Author Topic: SQL / SQLite for Beginners - Tools, Tips, Tricks, and Traps  (Read 7998 times)

ital2

  • Member
  • Joined in 2017
  • **
  • default avatar
  • Posts: 115
    • View Profile
    • Donate to Member
EDIT June 5, 2017:
Title change and intro
(Previousoriginal titles: "SQLite, SQL, SQLite Expert", then "SQLite, SQL, SQLite "Expert" (ha, ha)", for lack of warnings when stacking up database updates which in combinations can get dangerous and which you would do one by one if you did them by hand)

This thread does NOT intend to replicate the usual sql intros, but wishes to give additional practical advice for beginners those intros' authors did not think about.


Original post:

First, some quite obscure trap for beginners, stackoverflow helped me to find that out.

For many a command, not using quotes (single or double) will work fine, so you will not discover it's faulty syntax.

for example
select * from tablename where somecolumnname = somestring
will work fine as long as there is no column (another column) which is called/homonym to somestring

example:
select * from A where T = V
will work fine and display all records where there is a "V" in column "T"
IF there is not another column which by any chance is called V

IF there is, SQLite will simply do not show the resulting records, no warning, no error message, nothing, so when you've got some other columns and your requests work fine in general, it'll be probably a lot of time before you realize SQLite will not display many wanted query results.

You know ask why would anybody call many columns by single characters, and also many string values by single characters, to begin with? That's right, but I had just a few 1-character columns, incl. "C" for Code, with 1-character codes in it, for several, different todo-categories, just 6 or 8 characters in all, all single ones, and so I wanted to have it neat and without sacrifying space, and so I called that "Code" column "C".

This works fine, except for my code "h" in column "c", since I also had another 1-single column called "h", and without the quotes, SQLite is unable to differentiate column names from values if they are identical, even if you would have thought that the "where c = h", by its syntax, must have told SQLite the first is a column and the second is a value, but no.

Thus, it's absolutely necessary to write in the official syntax

where "columnname" = 'value'
which is double quotes for identifiers and single quotes for strings/literals
(if at least it was the other way round, I could memorize that better...)

in order to avoid lots of problems which, as the above use case has shown, are far from obvious and could have you get incomplete results without even discovering the problem.


EDIT June 1, 2017: Title change; see next post


EDIT / Add-on June 5, 2017: Null
The web abounds of discussions if you should use null values or empty strings; I won't add to add to that discussion but rather want to indicate some more beginners' traps caused around Null/empty values; btw it's of interest that Oracle's and FileMaker's sql flavors which both try to do away with the Null problem both don't succeed at it, so switching over to them will not allow you to ignore what it's all about.

When I imported my csv data, I did that, as I said further down in this thread, in SQLiteBrowser 3/DB Browser for SQLite (they don't know how to name their tool, it's not 2 different tools here), then opened my databases in SLite "Expert" (see next post), since the latter's "Personal" edition doesn't allow for import or export (but you can do some select, then ^a, ^c, then treat the resulting csv clipboard by macros, for example for printing or for exporting the data into anything else).

So the first problem described here may result from this 2-step setup. Anyway, blank fields from the imported data were zero-length/empty strings, while the same for records I then created in SQLite "Expert" (fields I simply left blank, in newly created records) were/are Nulls. If you don't know databases at all, you will think that's the same thing, database experts will laugh about what I describe, but for database beginners, it's important information.

This also implies two things: You should start your database experiences with data you intimately know, and/or you should always set the option "display Nulls" to "on"; both measures will help with at least becoming aware there are problems. Since, and that's the problem, your selects will not display the expected results if you don't  always cope with the " where x = (or <>) '' [these are two single quotes, not one double one!] or where x is (not) null" problem you will have got now.

Also, traditional logic is different from SQL logic, and you must become aware of that phenomenon, too. To start with, we use the (possible) column setting "No Nulls allowed here" for "in no record, this field must be left empty", but in the SQL 3-tier-boolean world, this just means "absence of Nulls", not of empty strings; similar for boolean fields (yes, no, null (allowed or not)) and for numeric fields (some number, null, or, very bad in most cases, some "special" value for "no number here" from developers wanting to avoid Nulls at all (and in case very high) cost.

Also, you should google for "not in vs not exists", and even when you think you are aware of the problems and do it all correctly, you probably get wrong query results again because of your having left out the fact that Null is NOT the opposite of "True"/"Yes", and that this logical sql truth extends to any field type in sql.

Many authors insist on the fact that Null stands for "unknown"; since we also and even mainly use it for "no value"/"field empty", and the meaning of "unknown" is rare, we may overlook the fact that as for the sql logic, these authors are perfectly right. And we fall into the trap of writing "where x <> y" instead of writing "where x <> y or x is null": In our logic, Nulls are excluded anyway there, in sql's logic - which we must observe when we want to get correct query results -, it's "who knows about these Nulls?" instead, so they have to be mentioned separately in your query in order to deliver correct results.

Somebody in the web wrote something along the lines of, "Which consumers/non-professionals still access data directly, nowadays?" - well, it's the traps in sql which discourage people to delve into sql, since they prevent them from getting the expected results. It's correct that you would have to delve deep into sql in order to gain some real expertise in it, but some advanced facts must be known to the beginner, too, from day one; it's not like playing the piano where you safely attack some piece of Kinderszenen without needing to know anything about the intricacies in Kreisleriana.
« Last Edit: June 05, 2017, 09:25 AM by ital2 »

Tuxman

  • Supporting Member
  • Joined in 2006
  • **
  • Posts: 2,507
    • View Profile
    • Donate to Member
Re: SQLite, SQL, SQLite Expert
« Reply #1 on: May 01, 2017, 11:06 AM »
Was that a question?  :huh:

widgewunner

  • Member
  • Joined in 2009
  • **
  • Posts: 93
    • View Profile
    • Donate to Member
Re: SQLite, SQL, SQLite Expert
« Reply #2 on: May 01, 2017, 11:14 AM »
Thanks for the public service announcement! - I am new to SQL and am using sqlite2 in a couple small projects and this info is indeed useful/helpful to me.

Thanks again!

ital2

  • Member
  • Joined in 2017
  • **
  • default avatar
  • Posts: 115
    • View Profile
    • Donate to Member
Re: SQLite, SQL, SQLite "Expert" (ha, ha)
« Reply #3 on: May 01, 2017, 11:33 AM »
Then, a word on SQLite Expert Personal

As I said in the "Navicat Warning" thread, I recommend this free software, it's really pretty. It has got some bugs, but not of the data-destroying kind, just minor window problems and such.

I said that I use it on my pc, but I left out an important info: Since it's the free version of a paid program, it will not import data, so as it is, it would have been unusable for what I'm trying to do, translating some organized text files (and also csv-files from ListPro) into databases / flat database tables (flat to begin with, looking out for doing it perhaps more to the regulations later on if I find the time).

So I use SQLite Browser / DB Browser for SQLite (free, they don't know how to name it, anyway, for XP I use version 3 of it since 4 will not work there): Import from csv-file into table, then save the new database. All other things I then do in SQLite Expert Personal. csv means comma-separated, but obviously a comma as field separator isn't the best alternative, so I use tab-separated, but it's with the .csv suffix anyway.

Since in SQLite Browser / DB Browser for SQLite, I create the database, for example cds.db (from cds.cvs), I can then open my CDs-database within SQLite Expert where I format the columns, and so on.

Now my explanation why I just don't buy SQLite Expert since I really like it a lot. Well, my psychological loss is too big. Last year in summer, I had already looked out, superficially, for SQLite frontends, since my problem of having data in text file, as explained in another thread, is a persisting one, and then I already had had the idea to do something about it.

It was 60$ lifetime, which was an absolute treat, and I promised myself to buy that fine program as soon as I would need it. Now it's 100$ with 1 year of updates, which means instead of costing me 60$ plus VAT, this same program (btw, the price rise came for the "old" version, not even a new major version in-between) would cost me hundreds of $$ over the years. Of course it's all my fault, and I don't blame the developer, but since I lost the lifetime option, now paying a price almost double of that one, for just the current version, plus update costs again and again - it's unacceptable for me at least for the time being where SQLite Expert is a fine SQLite data browser (I like it a lot more than SQLite Maestro, same price, so you see where the hike had its idea from), but where for example there is no graphic designer, let alone that imaginary piece of joy I've just described in the "Navicat Warning" thread, and there are no stored queries yet, which are available - as a graphic designer, but not a good one - in SQLite Maestro, but as said, stored queries in SQLite Maestro cannot be triggered from the tree entries but must then be triggered by an unnecessary intermediate step.

SQLite Expert is in continuous development (got just another minor upgrade), so it's very evident it will get at least some of its missing functionality later on, and with no doubt, people who didn't come too late for the life updates, as I did, will have got an absolute treat. (It's been very unfortunate that the price hike and the dismissal of the lifetime license came both at the same time - together with the almost 50 p.c. rise dollar vs euro, that all combined is not too bad for a developer who's in the euro zone...


EDIT May 25, 2017
You will find more info upon SQLite Expert, also in comparison with some competitors, in my May 25, 2017 add-on in my thread about trial conception ( https://www.donation...ex.php?topic=43835.0 ).


EDIT June 1, 2017 - SECOND Edit for the day
Destroy Your Data with SQLite "Expert" (ha, ha)
You will be interested in naming your columns in some standardized way, between your different databases, in order to greatly facilitate your management of pre-defined sql queries, so for example you will use C1...C4 (see my edit of the post below) for (possible) hierarchy, and even when you only got one code for the time being, that column should be called C1, not just C, in order to standardize your sql selects as far as possible. Also with "Titles", "Items" and such, just name it "T".

The same with "Authors" and "Artists" and such bad column names, just call that N for Name, and the Christian/First name should be FN; if you call these columns N1 and N2, you will have the logical problem that when your sort, you'll either sort by N2 first, then by N1, or you will have those columns in the wrong order on your screen (if you want to have the traditional set up with first name first, then family name).

So I lately did a lot of column renames, in order to standardize my tables / databases, and my selects - they all are somewhat different, but they should identical in all parts similar.

Then, in one of my tables, I made a big mistake, with SQLite "Expert" (ha, ha), by switching the names of two columns, then triggered "Apply": This, as you will already have guessed, immediately deleted thousands of names "N", putting the FN into N, and leaving the FN column blank; of course, this happened without the slightest warning.

My solution: I had the Ron's Editor csv file from yesterday or the day before, so I dumped a copy of that (as text) into Beyond Compare, the later-on-deleted column set to the end of the bunch in order to be able to visually compare in BC; I also dumped the remaining database into a csv and put that into BC, again as text (BC data compare may function with additional settings; out of the box it did NOT align the datasets); then from the differ I updated the the original csv from the partly destroyed newer one (not in BC but in Ron's Editor), then with the updated original csv I built up the new database: Two hours for nothing, but imagine the loss if I hadn't had a recent csv of my data available.

What do we learn from this: Do daily database backups. Do an additional backup before doing any database re-"design". Don't assume software which calls itself an "expert", is an expert, but assume it probably misses relevant security code analyzing the possible interaction of several steps which will be done from what it lets you "combine" in some task list*, and by which it will possibly destroy your data without even knowing.

You know the principle: switching names a and b means renaming a to c, then b to a, then c to b; SQLite "Expert" (ha, ha) obviously thought it could make without the additional step. If I had done the renames "by hand", by sql renames,
- I would have made a backup first, since I would have trusted myself less then I obviously trusted this "Expert" program, and
- I would have made the right steps in the right order since I'm aware of the problem.

In SQLite "Expert" (ha, ha) I relied on the software to make those steps, in the right order, and which failed, but I made my choices within some tasklist where you can ask for all sorts of changes, and then the program, upon "Apply", will execute these tasks one after another, it seems, and where the SQLite engine itself will not stop the execution, by its own internal safety routines, it's obvious this program proceeds, without taking account possible unwanted interactions between the various commands in the line: There is no safety procedure implemented BEFORE sending the commands to the engine, like you would do in your head, by doing it manually.

It would be of interest to send such requests from within other SQL frontends, in order to check if they come with such checking routines or don't either. This comparison would only be valid for different frontends but the same database format (here SQLite), since, as said, it's perfectly possible that some SQL engines have some of such routines in-built so that any frontend applied upon them would work fine, either by applying the necessary additional steps or by rejecting the commands it cannot faultlessly process in that context.

And what can developers learn from this? FIRST exterminate such fatal bugs (here: not checking if any rename was in fact a switch: recursion probe: if the commands are sent one by one to the database engine, how could that discover such a problem? So it must be done by the frontend, which also triggers the necessary intermediate steps/commands.), THEN triple/quadruple your price if you really think you must do so (here: from 60$ with lifetime updates to 100$ with 1 year's updates).
« Last Edit: June 01, 2017, 04:56 PM by ital2 »

ital2

  • Member
  • Joined in 2017
  • **
  • default avatar
  • Posts: 115
    • View Profile
    • Donate to Member
Re: SQLite, SQL...
« Reply #4 on: May 01, 2017, 12:44 PM »
Here: The "like" query for strings as a real alternative for not-too-extensive databases

classic sql selection:
select * from tablename where "x" = 'y' -- display all records where column x has got the value y

"like" syntax:
select * from tablename where "x" like '%y%' -- ditto for the string y anywhere within the value in column x

It's known that simple databases - we're not speaking of full-text search here, which is available in SQLite - are not perfectly suited for text search, the "like" clause is said very slow. This certainly is true for big databases, but for thousands of records on an old xp pc, even "like" will bring the results in less than a second, so it's perfectly adequate.

To give an example, my list of CDs:

ID (SQLite ID number)

Code/Category

Range (bear in mind CDs are physical stuff, so they only do have one place if not copied, no physical CD in more than one physical range, so the "where to find" info is not redundant even though most of the time (only), Category and Range are identical; in pop, categories are by countries and then in case decenniums; in jazz, Category is the instrument of the main artist),

Artist (can be several, for jazz in particular; it's just my classical music CDs that I have got in a different list, the main difference being the fact that the composer column is the main field over there, where pop, jazz, and so on are sorted by artist)

Year (not of a single CD but year of birth of the artist or year of creation of the group)

Titles (which means different CDs, not titles on a given CD, that info being on the sleeve; it's just in certain cases that I put exceptional titles in parentheses here, in order to know where to look for them; also the year of the CD is in parentheses if I took the time to look it up from the sleeve or from elsewhere)

From the above it becomes evident that my data isn't in perfect database format since as said, some records can have several artists, and it's similar for the titles: I put some solo artists together with their respective (main) groups, and then, in the titles, I put some additional info; it's evident I need (and have) links.

Instead of lengthy explanations, I just put an example here, with the query and its immediate result; from this it becomes clear that a simple (and not too big) SQLite database, with systematic "like" instead of "=" query, is also suitable for text data which is in a less-than-perfect form, and where to bring it into such perfect form would not make too much sense, because of too much work, for not really much more effect in practice (left out fields without interest here):

select * from cds where "artists" like '%stefani%' or "titles" like '%stefani%'

1)
[Artist:] No Doubt
[Category:] A8 [this means Pop, USA, Eighties]
[Year:] 87
[Titles:] (Gwen Stefani 1968, Eric Stefani) a) Rock Steady, b) Gwen Stefani: The Sweet Escape (2006), c) Gwen Stefani: Love. Angel. Music. Baby. (2004)      

2)
[Artist:] (Gwen Stefani see No Doubt )
0 [default when not applicable]
0 [default when not applicable]
[Titles: empty]

From the above, it becomes evident that it would be very welcome to have some input device (macro tool) which would allow for typing the search string just once and then put it into both query statements. The same macro would also apply to queries for titles, or, for speed reasons, you would apply a second macro, with just one query statement (the string just being searched in titles).

A macro may be easy on the pc, it should also be available for Android and/oder iOs/iPad, I hope. Here again, if you make the slightest typo, most of the time you will just not see the wanted results, without any error dialog, for example if you write "artist" instead of "artists", so doing this by some macro tool is not only a question of convenience, but of avoiding typos which may leave out wanted records.

Also, if I put my links to artists from the "artists" column into the "titles" column (which should be quite easy), a simple query would be sufficient in these searches for an artist also belonging to some group, but it would not cover cases where, as said mainly in jazz, there are main, and "secondary" artists, the "secondary" being just the ones by which the CD in question is not filed. As you can imagine, in jazz, it's not so neat at all, but in my text files, every important artist is either named in the artist or in the titles field, so that with a query over both columns, they will be retrieved reliably.

In other words, in many practical, private or office, use cases of not-too-extensive data collections, "like" is the solution to use a light database instead of a text file: on a modern pc, some 20- or 30,000 records should deliver almost instantaneous results, and that without enormous adaptation work beforehand which could often not be justified in terms of time investment.

P.S.

Tuxman: Any statement is an implicit question, too, for people who'd be able to add significant info/advice.
widgewunner: Thank you!

EDIT:
Just imagine music group names with or without the leading "The": Who; The Who; Who, The? If you all made it neat instead, that and similar ones will be additional problems; ditto for names in an office environment (but where most of the time the databases are very big and thus name search should be made possible by "="):
- prefix before first name
- first name
- prefix before last name
- last name
- suffix after last name
Not speaking of the multiple variants, according to country, too, and then even according to existence/absence/combination of these elements, how these elements are then to be combined differently (!) in the address field and in the salutation.


Edit June 1, 2017
Writing from some more experience with transposition from text data to database data:

As said, you'll need a RegEx editor, and you'll need to know how to do RegEx replace with it, and then you'll be able to achieve about 80-90 p.c. of the necessary translation work by automated means. It's also necessary to have a RegEx editor which allows for details, for example for only replacing the very first ":" in each line, instead of further ones, too; this avoids a lot of manual work after visual check.

As for the result in the database, you will want the family name of the very FIRST author/artist/etc. of probably several ones as the START of your author(s)/artist(s) field since you will want, for example for print-outs, to be able to SORT by them ("order by..."); for just FINDING authors/artists/etc., you'll do as explained above ("like" instead of "="). Thus, you will need to do some macroing in order to seperate the first from the last names, which means in practice that you clip by the last space (before the first comma if there is a comma) in the field, except when there is some "di/von/de/of", by this allowing for more than one first (or middle) names.

For example, when there is an institution as the author, this will not work, so you will have to do some visual checking afterwards, for this like for everything else, but here again, it's a very good idea to combine several tools; for example, visual checking is much easier within a csv editor (I use Ron's Editor for that, but do NOT try to SORT with that, it's bug-ridden in this respect but the developer is not interested in de-bugging it; at least he does not react to mentions of those bugs) or even the target database, but the editing should be done by RegEx, within your editor or such (2-screen or big-screen setup): At this stage, you'll use the database just as a visual checking tool but you work on your editor csv file, which means you'll create the same database several times, up to it being almost perfect, then only NOT going back again to other tools for editing.

Before the final version, all your fields will be in varchar format and without size limitations, since all too often, depending on your original text file and/or your editing ("|" or something else as field separators) in order to distribute your data into the respective target fields, SOME contents of for example fields 6 and 7 will be within fields 7 and 8 instead, and since thus things you can quickly rectify in a csv editor (at least in Ron's Editor), you'll do your editing in some systematic way, sometimes even doing changes in the csv editor, then saving it for further editing in the text editor for RegEx, and so on - mixing things up will end up by inadvertantly discarding corrections already done.

Instead of buying a csv editor and/or learning RegEx, you probably can do a lot of things within Microsoft's Excel, but you'll probably get a double-quotes problem which can be avoided otherwise; anyway, you should replace all double-quotes with single-quotes in your original file(s) to begin with, before getting to any editing in the copied version of the original file. Btw, since your editing will probably have unwanted effects sometimes, you should do some sort of versioning (between your visual checks in the csv editor or in Excel or in the database), and of course you should store your original source file in a safe place as they say.

As a general rule: If your editing had unwanted effects on SOME records / datasets / text lines, rectify them manually at once, before doing other automated editings; if they had such unwanted effects on a lot of records, though, go back and devise some other (/ some other RegEx) strategy.

And: Many users don't use text files for data storage but Excel files to begin with; it's evident that they then will get all the advantages of a database (SQL queries, incl. standardized ones and ones for filtering), even out of their flat files (before, or even without ever normalizing them), without all the semi-manual fuss described above, so to store half-way standardized files in Excel instead of in text files seems to be a very good idea: The field format is common to spreadsheets and databases and thus forces you to a much better standardization from start on then the text format does, so you'll get much less "micro-architectural" variants than if you use text files, even if you had thought you had been entered your data in a standardized format in there.

As for "sorting fields", I mean fields FOR sorting here, it's evident you can easily establish a flat tree structure, by (creating, and then sorting by) fields like C1/2/3/4, for Code1..., for categories and possible sub-categories, while the latter can remain empty when not needed.

But as implied above, databases, even "primitive" flat ones (and Excel, for such things), only get "interesting" when you're able to trigger technically complicated sql (or other) selects very easily; in other words, you need 1-key(-combinations') "stored searches", but with variables for the key terms; it's surprising that current database frontends' developers didn't grasp that so that you have to write your own macros for it.


Edit June 2, 2017
Printing ("Reports")

I didn't look into the various "Reports" facilities of paid/trial frontends; from previous experience, I would assume most of them are not very flexible so that you must accept what you get; for example, from Navicat, you can buy some extra "Report Generator" so that it's safe to assume the report functionality of the Navicat frontends is very basic; also, there is some famous generic report generator I don't actually remember the name of and which isn't cheap either.

Thus, printing from some database frontend does not seem to be easy, but it is, with some macroing. First, you must identify how you want to get the records; for example, you want the first record separator ("|" or other) of each record to be translated to ": ", the second one to " - ", the third one to a ":" but with a code for printing bold, the fourth one to be translated to a comma and a code for end of printing bold, the fifth one should be a code for newline, etc. - it goes without saying that we're speaking of database data and thus we're speaking of perfect standardization of the records.

I've spoken of this problem before, you need a RegEx implementation which allows for differently treating the first, second, third... occurence of your field separator in each line. In practice, this means you need an implementation which allows for treating each line individually, and then, instead of trying bombastic macro code, you just set up specific RegEx commands for the specific field separators 1, 2, 3..., and you run the first RegEx to the first occurrence of the "|" (or whatever) in each line, then the second RegEx to the first (!) occurrence... and so on: Since each RegEx replaces the first "|" in each line, the next run will automatically replace the next occurrence.

So you must find a RegEx implementation which allows for treating each line individually, but in bulk of course, and which allows for treating just ONE occurrence in each hit region (here, the lines); this will unfortunately exclude some of the available editors and their RegEx Replace implementations, but then, you will want to have the whole macro available for any printing anyway, instead of manually entering the different RegEx commands.

After this, you will have your csv file, exported from the database file, and you will have a text file which originally will have been a copy of said csv file and which is now a plain text file, in case with markup codes for formatting.


Oh, I forgot: You will need titles and sub titles for prettier print-outs, and you also will want to have automated page breaks. You will run an sql select with "group by" (without the further data of course, in our example above without "N", "FN" and "T" and so on), which will give you a list of the groups, and then you put that list, after the necessary formatting, as bulk into your database (insert into ... (field, field...) values (v, v, v,), (v, v, v)...; - you know how it's done): This will give you sort of "empty" records but which are perfect titles and sub titles even for browsing query results on screen (but then without leading blank lines, while for printing, you can easily insert blank lines or page breaks before those titles/sub titles). In the future, you will not create new categories or sub categories without creating the respective "blank" record, the record with just the categories, first, and that's not too difficult to remember, since:

You shouldn't create new records entirely by hand anyway but create a macro which allows for creating a new record as a copy of the current record, and if you systematically do this, this means that you will create new records by first going onto the "title" / "sub-title" = the "dummy" category record, trigger the macro, and then just fill out the real data; this way, it's technically impossible to ever forget to create the dummy, "title", record for some new category.

Oh, and I forgot, how to then automatically insert the codes for leading blank lines or leading page breaks? You will insert an additional column, name it "pc" for "print codes" or whatever, 1-char numeric, default 0, blank line before = 1, page break before = 2, or whatever, OR you could write a macro which afterwards checks the blank content fields, but that would be less flexible since it's not taken that you want a blank line before ANY record where C3 and C4 are blank (lesser group change), and a page break before ANY record where C2, C3 and C4 are blank (and all the rest, of course, since we only consider titles of any level), so an individual setting is probably best (and will "cost" 1 digit per record). (Of course, your macro mentioned above will then also translate the "1" or "2" into the needed formatting code.)


So, you will have got some select result which then is perfectly pre-formatted by your macro but which looks horrible to the eye. What to do with such a thing? You can put it into InDesign or some other dtp tool, but for most use cases, that would be way too sharp an effort. It's probably possible to import it into Microsoft Word or such, but the question remains, how much formatting will be preserved, how much of it will be lost? It would be ideal if there was some pdf printer (tool) which could read some formatting at least, but in fact, I didn't even find any which preserves page brakes (ascii 12 for example), so let alone for bolding the content of some field content...

Thus, you may think of html, which should be perfectly possible, but then, printing from html is not something I had never considered to do. On the other hand, it's very easy to enlarge your macro just that little bit as to add a little rtf header and the closing brace, and running a macro translating all possible special characters (à, ù and so on) into the respective, 4-chars rtf codes, is easy since the web abounds of the respective tables for that task.

Then, you open your newly-created rtf file from within an rtf editor (for example WordTabs, free) and trigger "Print" - that's it, or perhaps it's even possible to send an rtf file directly to a printer, the necessary commands (paper size, margins, and so on) should all be available within the rtf language.

I'm just assuming here, but let me guess that the above work-flow (which is entirely free) is much more flexible than some expensive database "report" programs are, at least as no graphics are involved, and I admit that I didn't bother to look up Tibetian-to-rtf, but for the rest (and by printing from WordTabs or Works), it works fine already... to be refined of course, and when I have time, I'd like to get rid of the additional rtf-editor-for-printing step - all of rtf is available on the web, but it's quite some stuff to wade thru, for direct printing.


Oh, and for standardization, all the tables in my one-table databases are called "main"; helps big with writing heap select-macros.


Edit June 3, 2017
To clarify why RegEx here.
I'm not speaking of the classic RegExReplace "somethinga fieldseparator somethingb fieldseparator somethingc" by "somethingb fieldseparator somethingc fieldseparator somethinga", for which RegEx is needed in order to maintain the original somethings to be relocated (you'll work with parentheses for groups and numbered placeholders for them), it's just needed here in order to treat the lines (records) one by one and then applying the "replace once only" (in the specific way of the respective replace of the replace stack) to every LINE. You probably can do this by other means (for example have a look at KEdit, its price has come down from 129 to 99$ recently and which possibly could do it, did not check), but for me, doing it with RegEx replace has proven simplest.

Instead, you could write a one-line classic ReEx replaces "doing" each line/record for all which is needed, then (with the same command line) the next, and so on, but I suppose that will take a little bit more execution time (by the multiple replaces 1-by-1 of the field contents (which are left alone in the strategy described above, and which I prefer for that reason of "minimal invasiveness" then), but it's correct that it may be simpler to maintain specific 1-line replaces, one for each database select-result file (independently of course if this result comes from flat tables or from a more or less normalized database), than doing multi-line replace commands, one line for each field separator in the record. But then, I prefer the latter also for the fact that in a multi-line command, you will get neat comments at each line, whereas in a 1-line RegEx, you would need to write down the respective comments not as near the respective command but for example as a multi-line block below, each line referencing one group in the 1-line replace command, so at the end, the 1-line flavor is not as elegant anymore when you put down comments, and it's evident that even when you do, changes aren't as quick to do as with multi-line since you'll always need to check if your change concerns the right element (and not possibly its left or right neighbor, inadvertently), while in the multi-flavor, with the (right) comment in the same line, it's always evident where you are and do your changes. But that's a discussion of RegEx styling, both ways are perfectly possible.


Oh, and I would like to modify my claim above that the described strategy may be "more" (?) flexible than out-of-the-box "report designers". For example, it's easy with the former to automatically insert a leading blank line whenever the first, or even the second character of some field changes; this would be helpful for example for the field "A" when you set changes of the fields "C1" to "C4" to "leading page break", and would not even take much white space in case you print in more than 1 column - but good specimens of the latter variety should do that too, of course, so let's say the way described here is AS flexible as good "report designers" (very probably) are.

You just run another macro element which retrieves the content of that field in every line/record, compares it with the stored content of the previous one, and if character 1 or 2 isn't the same anymore, it changes the content of field "pc" ("print code") of the line from 0 to 1 (if the one of the previous line has been 0, not 1, 2 or other, before, so this will not affect records after titles/sub titles, or you do it by "changes, but no changes from nothing", titles'/sub title's "N" fields being empty, in our example above); or you simply bolden the content of the "N" field in the output upon such character changes only, instead of boldening it for every record - as you see, the possibilities of revamping core database outputs are seemingly endless, just by running text macros on the csv, in the right order.


As said above, you will want titles/sub-titles on paper, and on screen, and on top of the respective real data records, not below them, so with the above, you will have an order problem:
select * from main order by C1, C2, C3, C4, B, N, FN, T -- I renamed "PC" to "B" for "breaks"
does it wrong,
select * from main order by C1 asc, C2 asc, C3 asc, C4 asc, B desc, N asc, FN asc, T asc
is ugly, so, in order for the real select
select C2, C3, C4, FN, N, T from main order by C1, C2, C3, C4, B, N, FN, T
to work correctly,
you either code line breaks as -1, page breaks as -2, but I don't like negative values when not needed, or you set the "B" column's default value to 3 instead of 0 and set it for line breaks to 2 and for page breaks to 1; this will even leave you the 0 for a possible other code before you'll have to do an "update main set... where...". I concede a default of 3 isn't pretty but then you don't even have to display it (see the last select); for new records, your macro will put it in automatically, and sql lets you sort ("order") by invisible columns.


We're continuing the "refine it" part here: When a field is empty, you will probably want to get another replace string as the one you want for a non-empty field, for example if you have 4 hierarchical categories C1...C4, you want them "C1 - C2 - C3 - C4: ..." but when subcategory C4 is empty, you want it to be "C1 - C2 - C3: ..." instead. How to do it, since RegEx does not offer some "if it's the full string (with variable part here, that's a given), replace it by (the various part and) "a", and if the variable part is missing, replace it by "b"; look-arounds do not help either if you always replace the very first string of the kind of each line, in every iteration, and a non-replace would create chaos, any un-replaced field separator wrongly becoming the first one then.

So in most cases the replace would be done, which would result, in our example, in a "C1 - C2 - C3 - : ...", and later, you simply would replace any " - :" or even " -  - :" by a ":". If that's not possible because of possible occurrences of these strings in the regular text, you simply replace all field separators with the wanted strings but in which you maintain the original field separator, by replacing it with a special character which does not occur anywhere. For example, if the field separator is a tab, do the replaces as wanted, but replacing the tabs by an additional "|", too, which gives, in our example, "C1| - C2| - C3| - [C4 being empty]|: ..." = "C1| - C2| - C3| - |: ...", and then, you'll simply delete (always by macro since you'll know beforehand which strings can occur by this) unwanted "|" combinations (here: "| - |"), and in the end, all remaining (single) "|"; Perl is said to be best, so perhaps it can be done in simpler ways in that language. Anyway, you clearly see that you don't need to just use spaces, then delete the multiple ones, but you can do anything you want; let's hope the commercial offerings go into these depths, but then, 2/3 of their development efforts inevitably go into the GUIs.


EDIT June 12, 2017: Some (useful and/or very important) Details

Output: RegEx / Replace

To my surprise, I didn't find any RegEx editor which allows for only replacing the very first occurrence of some given (sub-)string in each line, so I had to do it the opposite way described above, by changing every line in one occasion, the whole line, line after line, and this gave those horrible 1-line, inscrutable RegEx replace code I referred to above. As you know, you must group the sub-strings by (), and in order for then not getting totally lost, I did a comment line, with numbers 1, 2, 3... 25... immediately above the code line, for every ()-group and with the necessary numbers of spaces before and between the numbers, in order to identify every group by its number - since in the replace part, you must refer to those numbers but which are not in the what'-so-be-replace part of the RegEx code - a RegEx flavor where you write the group numbers directly into that "original's" code would be very welcome.

The above applies to the regular, the "B=3" lines (and thus, the "B=3" is checked within this code line and will thus leave alone lines where this condition is not met); you'll do your special RegEx replace code for other lines, for example for B=1 or B=2, for other conditions, and you'll also don't necessarily replace the original line by the final rtf code but for some formatting code, you'll use placeholders (for example little/suspended/superscript 1/2/3/a and such), and you'll do their final replacement later on (incl. extermination of unwanted code combinations, see above): This will greatly enhance the clarity of your code (so you'll not be get lost in it) AND will greatly help you with formatting changes you will want to implement afterwards, be it that you want some else formatting or be it as alternatives for several different output situations.

Also, you must pay attention to a correct, AND to a smart, order in your replace code (which replaces will be made before others; if you do this smartly, you can greatly reduce complications in your code (differentiation of which lines / sub-strings are affected further on: "discard" anything which is easy to "discard" at some stage, then do the "complicated" things, then again take the "discarded" lines and do further replacements there, in case); also, only the core replacements where RegEx is really needed (for groups) are done by RegEx, most other replacements are done with regular replace commands.

In order to not get lost in this bulk of code, read your text body into the clipboard, then have your code work onto the clipboard, and put a stop code after the first line, then the second one, and so on, and have your code display some of your (changed) text body to check if the replacements made up to the stop code are what you will have expected.

If you do the replace according to these lines of advice, it will all work out to your entire satisfaction, as it did for me, and your replace/rtf-formatting code will differ only in some core parts from one database to another; for almost anything, it'll be perfectly "reusable" - in practice, of course, you'll use the same code block for everything, doing just some "if database = a/b/c..." branchings where the code differs.

(You'll do a lot of replacements in order to refine your code, for example you will print/export only the very first words in bold, up to the first comma if there is any, of the content of the field "N", which means the family name of the FIRST author will be in bold, not further authors (in that book example)*, but if the author is some institution with several words in its name, all of these will be boldened; also, if you will have no first name (field "FN" is null), there will be no space (which would be inserted between a first name and a family name): All these details (fine-tuning) you'll realize by refining your code after the code works well in its core functionality, and placeholder characters greatly help with all this.)

*: Since I now have just one author in the "N" (for "name") field, OR the first author is before the very first comma in there (if there is any comma in that field), and the order within each sub-category (or overall, in order to get an ABC authors' list) is by these "first authors", it would be easy to put further authors (ie anything after a comma there) into additional, referring records, but since I do "like" searches anyway, I don't feel the need to do so yet.

Output: RTF

To my surprise, printing an rtf file isn't done easily, like you would send a pdf file to a printer, and I cannot do it do, there's too much special knowledge involved. So you will need some intermediate program in order to open your rtf file, and from which then you'll print the file. Forget the usual rtf editors, I tried some of them, like "Jarte" and others, and they will open your (correct) rtf file indeed, but then you'll lose much of your formatting, in other words, they aren't able to correctly read the rtf code of the file they will have opened.

Don't laugh and say, "then your code will be wrong", since in MS Word, it all works out perfectly, incl. for example printing-out in several columns, with title lines of several indentation levels starting a new section, so that's exactly the application I advise you to use for checking and printing your programmatically-created rtf files: First, Word is of great help to correct and refine your code, by checking what it will give in Word, and then, it's the perfect intermediate tool for printing as long as there is no basic tool available for sending rtf files to printer.

It goes without saying that such "raw-data-transformation" of database data into "pre-formatted target data" not only works with rtf code for printing, but also, with the respective html/css/js or mark-up codes, for web publishing or for further treatment in InDesign or any other (ex "desktop") publishing tool; even links to graphics or other elements, present in the database, could be integrated as needed into the target text body InDesign or similar then will work on. (Btw, the size of your database files will be extremely low, comparatively, since there is no formatting or other "noise" info within the database body, but just will be added programmatically after export into csv/text format; this doesn't apply to BLOBs, of course, but then, you probably will prefer links to files in the file system instead of BLOBs within the database anyway.)

To begin with, you will send the correct sql select statement to your database, in order to get the wanted data (only), in the wanted order; then, your text-processing code will do everything SQL isn't, or is less, able to do.

SQLite: rowid / ID / autoincrement / primary key

As for SQL specifics you must know beforehand, there's the SQLite particularity that by default, it re-uses (!!!!!) row IDs that will have become "free", by deletion of the row/record in question; it's obvious this can create total chaos in your database (while for flat book lists and such, that is not really such a problem in reality, you'll want to make even such a basic thing "future-proof"). In the web, there is also WRONG information upon this phenomenon available, for example http://www.sqlitetut...qlite-autoincrement/ pretends/implies by omission of clarification that such rowid re-uses will only occur when your SQLite rowids have reached the stunning number of 9,223,372,036,854,775,807: this is definitely wrong. Fact is, most of the time, SQLite uses new rowid numbers, not numbers of deleted rows/records, but there is no guarantee whatsoever for this, in my tries, when I added (single or multiple) new records, then deleted them, then created new ones, rowid numbers of these recently created-and-deleted were re-used; HOW SQLite decided upon re-use vs new number is beyond my knowledge.

Thus you will need some ID column replicating the internal rowid but with the "autoincrement" setting (which then, but then only (?), also works for the internal rowid, or in other words, I don't see a way to apply "autoincrement" directly to rowid, you seem to have to replicate the rowid in some ID column which then, while remaining a replication of the rowid, forces that rowid to autoincrement; there does not seem to be a direct setting.

They all say that for performance reasons, you should avoid autoincrement at all cost, but imagine you do some zettelkasten database, with cross-references to other records, and then SQLite re-uses rowids/IDs of deleted records for new ones?!

Ok, that's not a perfect example since when doing cross-referencing, you should prevent deletion of referenced records to begin with, but this implies technical cross-referencing (cross-references either only in special, referenced fields (foreign keys) or cross-references-in-text with additional such referencing and the overhead that comes with that, and which is not possible within a home-made, hobby application) while in real life, you would probably use non-technical-cross-references, ie not-monitored ones, just by copying the target's ID number into your text within the content fields (hopefully with follow-the-link formatting), so that these "inline" references to other records MUST be absolutely stable, which, if you use SQLite along the lines of their common advice (ie avoiding autoincrement IDs), is NOT the case.

I did not yet try how it works if you create new records programmatically, ie by executing your own sql, only, but in practice, you will probably create many or even almost all your new records from within your frontend, and doing so will create plenty of re-used rowid numbers at least in SQLite Expert, but I think it's not the fault of this frontend, this time. Of course, there is always the solution of avoiding record deletions, discarding records not being wanted anymore into some sort of "archives", by switching a boolean switch in each record (for example column "a" for "active: yes/no", with or without some "cleaning" of them within the "active" table and shifting them into some "archives" table(s) every now and then), but it's probably a good idea to have such "archives" indeed, AND to provide for the possibility to definitively eliminate records, even from these archives, and this would make reappear the problem, and overall, how would you really prevent ANY possible (and potentially problem-creating) (manual) record deletion from a hobby application? In other words, you couldn't use any out-of-the-box sql frontend anymore but would have to write your own, deletion-proof frontend. It's evident that corporate applications would not come with such problems since there, for compliance reasons, it's even necessary that no record will ever be deleted, instead of being moved into some archives, but then, SQLite isn't the right back-end for any group application to begin with. So we're back to "autoincrement". But:

In the different flavors of SQL, it's more or less difficult do database design changes after creating a database, or after creating a column, and in some cases, your only means is to create a new database, with the characteristics you want it to have from start on, and then fill it with the data from your original database (fortunately, this can be done by sql code), and in the case of unique IDs in SQLite, that's exactly your problem here if you will have created your database without creating the autoincremental ID column when creating the database, ie the command alter table main add ID not null integer primary key autoincrement will NOT work, because of the primary key constraint ( see, among others, http://www.sqlitetut.../sqlite-primary-key/ ). (So when creating your database and your very first (empty) table in SQLite browser (see above), first create the necessary ID column, THEN only do the "File - Import - From CSV" command in order to create the (basic) database from your csv data, before "formatting" the (other) fields in your frontend.)

SQL: Mix-up of nulls and of empty strings

This problem described above did not come from my using two different frontends, alternatively, for the same database(s), but from my using Ron's Editor in order to quickly and "manually" shift data from unwanted columns into the correct ones, after sloppy RegExes of my original text data (see above); in Ron's Editor you can shift manually the content of many fields into another column very quickly, so this tool comes very handy for putting "far-from-being-normalized" text data into a table (csv) format for spreadsheet/database use. But whenever I imported or re-imported my data from that csv editor into my database (ie whenever I rebuilt the database from the updated csv file), I didn't have nulls but empty strings in empty fields, while, as said above, for newly created records from within the database (frontend) itself, fields left blank on creation correctly got null values.

Thus, csv editing of database stuff should come to an end at some point, then you'll replace any empty strings by nulls, and from then on you will have only non-empty strings, and nulls, to cope with in your database and your sql code. (I largely prefer nulls over empty strings, and be it for the only reason that I can simply leave fields blank when creating new records from within the frontend, without creating by this a mix-up of nulls and empty strings*; you'll do how you like to do it; it's just the mix-up that's really unbearable - or you'll get very complicated code if you want to get correct results.)

*: If you programmatically create new records, you could write your code so as to create empty strings for fields "left blank", but if you create them by typing within the frontend's table view, those fields get set to null if you don't manually set them to '' one by one, and voila, you'll need the code for empty-string-nulls-mixup further on, so clearly avoidance of empty strings in favor of nulls is preferable; your mileage may vary.

SQL: Field names

The ID field can get any other name (but not any other numeric format than INTEGER), but as a general rule, my uppercase field names above are a big nuisance when entering them into the respective sql commands; I'll change them to (equally short) lowercase names, and you should use those from start on (Windows doesn't make any difference here between lower- and uppercase anyway).
« Last Edit: June 12, 2017, 08:04 AM by ital2 »

ital2

  • Member
  • Joined in 2017
  • **
  • default avatar
  • Posts: 115
    • View Profile
    • Donate to Member
Re: SQL / SQLite for Beginners - Tools, Tips, Tricks, and Traps
« Reply #5 on: June 12, 2017, 08:56 AM »
I suppose home-made consumer databases have become very exotic, most users probably using Excel-for-everything, nowadays.

But some years ago, there had been some Excel add-on, allowing SQL selects in Excel. I bought, then never used it. SQL in some real-database frontend is so much more powerful, and it's fun to then dissect your non-normalized main table into more than one, "real-database" tables. Should I repeat it here, that the terms of user and of consumer are not synonyms, and that playing around with stuff, instead of playing around with GUIs that incredibly limit you, is so much more fun? Of course, as soon as there is some VBA scripting involved in your interaction with Excel, that tool can become very powerful, but for non-number-crunching, even then I don't see any advantage, compared with smart database use, and I'm positive about the fact that for getting brilliant results, some database interaction is so much easier than to try to get similar results out of Excel, without programming involved.

To be continued, with as much practical value as can be provided to consumerism-averse non-professional users, and even if their number obviously doesn't get into the millions, it's quite surprising nothing of this kind has been available up to now, so google should finally grasp its relevance, no?


EDIT June 13, 2017: Character sets / Code pages / Text formats/encodings
I finally did NOT need my rtf character mapping table (accented or other special character > \'xx, xx being the respective hex value) for characters like ò, ù, etc.: I found that some ANSI-UTF-8 conversions (and probably even back or in other formats, clipboard = UTF-16?) are done automatically, to and from clipboard and sometimes upon importing from .csv to .db - hope for the best but be aware of possible problems. In most use cases, the following - incomplete - observations will be irrelevant then:

My original text lists are identified by my editor as "Western European (Windows)"; this is ISO-1252 (which "is" ANSI, see the Wikipedia article on "Windows-1252" for further precision), but unfortunately, as a non-professional, you're never sure about your charsets since even automatic translations can occur (see below for an example): You hope for the best, but since it worked smoothly in my case, I didn't even pay attention; so, in order to at least half-way correctly describe the workflow here, I've gone back to the separate steps and have tried to identify the way of the encodings: paying attention to them, and in case making needed conversions manually, can probably help when things do not go smoothly.

First step: from my original text-editor-text to clipboard
(EDIT June 17: as you see, here, too, I forgot to check the initial step of export from db browser to clipboard, then writing to the initial text file; see correction below);

then in the rtf editor (I named it above but also said the developer doesn't hasten to eliminate bugs - don't do any sorting within that editor, you'd destroy your file (in the sense of mixing up the records' fields with the ones of other records)! -, so I don't want to name it again and again):

(1) "create new file from clipboard": Special characters are preserved, in the "File Properties" you can read "Text Encoding: UTF-8 without BOM" (BOM is the signature identifying the text encoding, "without BOM" is identical to "without signature"), so the program obviously (see below) has made an automatic conversion upon reading from the clipboard.

(2) If then you change the setting to "ANSI", the displayed text will not change, the special characters continue to be preserved, but the csv editor obviously does a second conversion back to ANSI (see below).

(3) Or you do, in the csv editor, "Import file - from text document", and here, in the dialog, the format is clearly identified as "ANSI"; here again, the special characters are correctly displayed in the resulting csv table, but the csv editor (if you use another one, all this may be totally different of course) has not made a conversion.

In the 3 cases, I saved the csv-editor file in the ".csv" format and reloaded those files back within my text editor: (2) and (3) were identified as "Western European (Windows)" again, as before (which obviously is ANSI), while (1) was now identified as "UTF-8 (without signature)", as it had been in the csv editor, so that csv editor clearly made those conversions (and not wrong guesses or such).

So from the results in the csv editor - of course, this is an additional step which you don't necessarily need in your use cases, but its result shows where we need to pay attention - we have two different start situation for further processing, UTF-8 (1) or ANSI (2) (in this case ISO-1252); in both formats, (West European) special characters are preserved (up to now; and this might not be the case for non-European special characters, especially with ANSI).

Next step, in DB Browser for SQLite/DB Browser 3 (see above): New file, then Create - Cancel (ie don't create table, but see above for the necessary autoincrement ID field which would to be created exactly here), then File - Import - Table from csv file (which isn't greyed-out NOW anymore; THEN table name "main"; this being done for both csv, UTF8 vs ANSI):
- UTF file import > Encoding is UTF-8
- ANSI file import > Encoding is ISO-8859-1 now, according to DB Browser, special characters are preserved (if you change to UTF-8 or "other", there are not, change to UTF-16 there is nothing left); this ISO-8859-1 is a subset of ISO-1252 (which is not offered as a choice in DB Browser; so in the former there are less characters available than in the latter, but it's sufficient for West European special characters).

Then "browse data" in DB Browser: You see the identification "UTF-8" in both cases, at the right-bottom corner of the screen. So an automatic ANSI-to-UTF-8 conversion will be made upon import of the ANSI-csv file by the SQLite frontend if your csv file is then still in ANSI format (see the automatic conversation in the csv editor above but which isn't made in every case: reading from clipboard yes, reading from csv file no (but which then can be made "manually" ie by choosing from a dialog)).

So, in your database frontend, you'll probably get UTF-8 encoding for the database, even from ANSI; at least that's the case in my workflow, when doing the import from an ANSI .csv file to .db format, in DB Browser; other frontends may not do this conversion automatically upon import, so you should check in case or try to get UTF-8 text files to start from.

Then, afterwards, any text processing you'll do is possibly done upon the ANSI format again, though, since when I do the the ^a - ^c in the database and put the clipboard content in my editor, it says "Windows (Western Europe)" again, with no further precision, and when I put the clipboard content in the csv editor, it says "UTF-8" again but we know from the above that the csv editor automatically translates the formats when they come from the clipboard.

Googling "clipboard content ansi or utf-8" brings some finds which seem to indicate that the operating system itself does automatic clipboard conversions (and even to UTF-16, seemingly not needed in Europe), which probably differ from one Windows version to the other, and with which applications you use have to cope with, making right guesses/conversations or wrong ones, so just let's say let's hope for the best but we should be aware of possible problems in different Windows / software / language regions setups; my setup above, with no cyrillic and with Windows XP and Word 2003 worked flawlessly.

As for RTF, the difficulty lies in the existence of the different versions over time and the fact that help available in the web doesn't necessarily apply to the version your target application understands; as said above, most rtf "editors" are really, really bad at this, and had I Word 2016 available instead of Word 2003, even more codes would be correctly understood. Also, it seems that many paragraph formatting codes don't work if you put them into the style definitions (in the header), so that you must repeat them within each concerned paragraph, and bear in mind that rtf is not a markup language like html for exemple, so \par ... \pard does NOT correspond to some <p> ... </p>, to bring the most prominent/easily misleading example.


EDIT June 16: Import from csv AND getting the id column right
As said above, a) "DB Browser for SQLite/SQLiteBrowser 3" is free AND allows for creating a table from a csv file; as said above, b) you would like to create a correct id column, replicating the rowid column, even if you do NOT judge it necessary to have it autoincremental (but which I advise, see above)*.

*: In fact, getting the "id" of the CURRENT record in a frontend grid is very useful but, without an ID column, often not possible. For example, if you are in edit mode (and in fact, the difficulty is not to get into edit mode, but then out of it again...), in SQLite "Expert", two fast "Home" key pressings (ie the second pressing fast following the first one) will bring you to the very first field of the current record, and then a ^c will bring you the record's ID into the clipboard, from where you can put that ID to numerous brilliant uses, for example for replicating the "default" data of the current record for creating a new one (while "not null, default value = x" will only bring you default values identical for the whole table, not for specific GROUPS of records, so there is a big interest in this even for this simple use case).

Now how to combine a) and b), since a) will create the table without any id, and then, it seems to be impossible to add that id column to the existing SQLite table?

Do a) in SQLite Browser since it comes that handy, name the table "main2" or something; even let alone the default column names, or rename them to the names needed later on, but leave the default column formats alone, it's not worth the effort.

THEN, in SQLite Browser or any other frontend of your choice (for example SQLite "Expert", as mentioned above), do SQL commands, don't bother with what your frontend would, or would not or not correctly, do for you, design and re-design-wise; of course, have stored the commands / column names in some text file for easy access, no need to type it all again and again.

Then, your sql command would be, in an example of mine:

create table main (id INTEGER not null primary key autoincrement, b TINYINT not null, c1 VARCHAR not null, c2 VARCHAR not null, c3 VARCHAR, c4 VARCHAR, fn VARCHAR, n VARCHAR, t VARCHAR);

(Edit June 17: here I had missed the default: b TINYINT not null default 3 or for strings: fieldname not null default 'somestring')

remember the table main2 already exists, too, holding the raw data; and then:

insert into main (b, c1, c2, c3, c4, fn, n, t) select b, c1, c2, c3, c4, fn, n, t from main2;

as said just before, you see here that any rename of the default column names in main2 is futile, for the "from" part, you can just use the default names as well, in their natural order, and for the first, the insert, part, you could even use your target column names in a different order than on screen if you pay attention to not mix columns up;

then delete table main2;

then in case, do some other formatting you will have not thought about up to then; this is probably the most simple way to create a correct table for csv/text data; from there on, you then can split up the data into other, linked tables; there again, you will first create the table, by your own sql command, then transfer data from the main table as needed, again from within the sql command line, not by - reliable or unreliable - frontend specifics.

Oh, and the sql "report" generators I spoke of above, are of course Crystal Report and SSRS. The former was quite ubiquitous in its time, even Microsoft had some OEM version of it, they now have SSRS while Crystal Report is now available from (and only for?) SAP. Anyway, my individually-programmable rtf export described above makes me wonder how many corporate users really use some (inflexible at least in direct comparison?) boxed tool, and how many would rather script what they really need, also in the light of the fact that sql alone does not do so much except for getting the raw data in some order, and then you'll need some macroing anyway, so the question is if this incredible effort put into report generators' GUIs, and which limits possible output variants anyway, is justified. As I see it, it's like sql as seen above: doing your own commands is more reliable, more flexible AND less complicated than wading thru those (limiting) GUI's steps.


EDIT June 17, 2017: NO clipboard export from SQLite "Expert" free version (but from SQLite Browser)
Above, I said that that program's free version allowed for selecting all or a "select ..." result, then copy to clipboard (^a, ^c) from which then I processed it further in order to export/print.

I now am unable to do so, but since I had installed a new version some days ago, 4.2.0.765, I re-installed 4.2.0.739 in order to check if the developer had taken away that ^a functionality, but in both versions it just selects the content of the current column. I would have sworn I had done it as described above, but now I don't simply remember anymore, probably I was in error. In fact, I did all the above for just ONE of my databases/tables yet, doing the export once, to an editor file, and there I must then have deleted all the "" to be found in combination with tabs (unneeded because of the tabs as field separators), and my clipboard processing always was done then from retrieving the content of that intermediate text file (programmatically).

So it seems that I did the export from a "select" result with SQLite Browser and then just forget about that; it works as described, but not from SQLite "Expert" Lite/"Personal". (While it's true that SQLite Browser offers export to csv format, of the database or of the current table, both exports just work for the table(s) in its/their unsorted/unfiltered format, not for "select" results, so they are useless.)

I discovered this error of mine just today when trying to export/print a second table; of course, the developer of SQLite "Expert" ("Personal") is not to blame if he wants to encourage users to buy his program; it's just that at this point in time, there are quite a lot of GUI bugs left thru the different versions (last version 3, then at least 2 for version 4) which are quite annoying, so for buying, now at 100$ plus VAT, one would like to have the (very pretty) GUI to function a little bit smoother before making the investment.

And another error of mine regarding this program: There is a setting "Show edit buttons" which shows edit buttons in every field, and clicking on them will open an edit pane just for that field - see my description above of such an edit pane for all fields combined though when I had missed this one. On the other hand, you'd need such an additional edit field for editing of fields with "long" content only, and the appearance of the unneeded buttons on ANY field is really disturbing, so it would be so much better to amend editing within the field itself for longer texts, or if the used grid doesn't allow for this, to have appear the edit field automatically (upon mouseclick/F2/field activation by arrow only, of course) whenever the text in the field is probably (!) too long in order to be edited smoothly in-line - that functionality would be possible in any such grid component, since it would just imply a "check by cursor home-end: addditional text displayed?" / "current text too long for this field if it was 1-line-only?", since the core problem isn't any additional text then entered, but to identify the insertion point for that additional text (or for text to be deleted).

EDIT some minutes later: I had just succeeded in selecting a "block" in SQLite "Expert" with the mouse (left button down), when the GUI was NOT in edit mode (in order to leave edit mode, I regularly have to change to another table, then go back - I said the GUI is ridden with bugs...), but my try to mouse-select the whole "select" result as a block (left top corner to bottom right one) was not successful, and even mouse-selecting a block then did not function anymore - so probably my initial export, some weeks ago, HAD been from SQLite "Expert" as I had initially thought. While the core seems to be stable, the SQLite "Expert" GUI is both the prettiest one on the market, and quite unbearable (I could list half a dozen other quirks of it).
« Last Edit: June 17, 2017, 06:53 AM by ital2 »

ital2

  • Member
  • Joined in 2017
  • **
  • default avatar
  • Posts: 115
    • View Profile
    • Donate to Member
Re: SQL / SQLite for Beginners - Tools, Tips, Tricks, and Traps
« Reply #6 on: July 13, 2017, 02:08 PM »
SQLite "Expert"

I had said that editing a value ("cell") was very cumbersome with that tool if the value ("cell content") is broad; I had said there wasn't even an additional edit pane sor the specific cell; than I had corrected, there was such an edit pane. Lately, I even made use of that edit pane, and I had to discover that the aleatoric, automatic word wrap in that edit pane was then, after "save [the edit]", replicated as hard returns within the cell content - so it's there, but it's worthless since it sort of destroys the text it's deemed to edit.

There are lots of updates, but they are offered in a way that you don't retain the previous versions, since it's "download and install?", then going directly into some temporary folder, not into your downloads folder, so if ever the developer wants to do away with some current functionality of the free (or even of the paid) version, that's the way he will probably do it. In plain English: Rather click "No", and download the latest versions manually then if you think you'll need them. Considering there are so many quirks which remain, over a multitude of these minor updates, I'd be interested in knowing WHAT is amended by the latter, I never see any difference.

Navicat

I had mentioned the around-30-p.c. price increase; some days later only, they announced new major versions (number 12 instead of number 11) but I don't see real development in areas where versions 11 fail; this being said without trialling anew, just speaking from their advertisement.

Correct line drawing

I mentioned the fact that in visual representations, the lines between FKs (targets) and their respective sources (PK) aren't drawn correctly - obviously because that would imply quite some more coding; Devart Studio for ... does not seem to be any better than the rest; the only db frontend I discovered so far and which seems to draw the lines correctly, is DBVisualizer (179$ plus VAT); if somebody knows some other competitor which does this right, please name it.

Remember Access? or: DB frontends limit your possibilities

I said above that it's easy to create some little sql commands which you then enter into the command line of the frontend, in order to get any info you want (if you also use sub-queries filtering the data before further questioning it), the only difficulty residing in too much typing, hence my advice to use some little group of macros, with some text input box for the search term(s).

It would be as easy to do after your splitting up your one table into several tables: The sql commands get a little bit more "complicated" (well, not really, and you will find all what you need in the web, neatly explained), and the result will be as neat as before: one single grid containing the results, in the frontend.

Where frontends totally fail is when you need to create new records or update existing ones, records spread over several tables (well, technically it's datasets split up into several records, of course): Here, you would have to write quite extensive "macros" yourself since you would not only need one inputbox and the "surrounding" sql commands, but you'd need a whole "FORM", all created from your macro tool, and then distributing the data into the form, for the part you want to see there, and then reading the data from the form you will have entered or changed there, and distributing it into the respective records in the respective tables: That would be lots of scripting, individually for any such little "application", like a ToDo set, for example (and not considering the need for a calendar and such things).

But IF you script all this, you will NOT need a db frontend anymore (and all the less so you'll want to have your macro/script enter the commands and the data by the command line of the frontend then), but you will only need a "grid", in order to display the data, since you'll do all the assembly work anyway. (NB: Some frontends may have some "API" which allows for entering commands / data programmatically instead of putting text into a command line, but they are overkill as a grid anyway, not presenting any added value, all to the contrary.)

In other words, Do-it-yourself with some db frontend is only possible as long as you limit yourself to one single table, as in a spreadsheet, and which is shockingly bad db behavior: You'll endlessly repeat the same data again and again (that's why in my descriptions above I have used 1-to-2-char codes instead of English words: in order to at least limit the typing chore), with all the craziness which comes with that - it's right that in a book list, such "category" repetition, over-and-over, may be acceptable, while in some ToDo database, it becomes VERY cumbersome to repeat all the multiple categories again and again, for every new ToDo; I've said it above, with a correct ID for each (1-table-only-) record, you can identify the current record and easily have a macro write a simili-clone of that current record, ie a clone with the probably-identical values (cell contents) pre-filled in the new record, in case ready to be overwritten manually by you, within the frontend grid ("in-line editing"), and with the fields left empty which you will (probably) want to fill out yourself, so you CAN simplify your typing work a bit IF you start, for every new record, from a quite similar one: So your macro 1 would be: Display the records of the kind of which I want to create a new one; and your macro 2 would be: Make a reasonable almost-clone of the current record (which the frontend automatically displays then); in-between, you'd probably do some arrow-up-or-down, and after macro 2, you'd do the necessary adjustments/additions.

But it becomes very evident, by this, that db frontends (just like Excel does) retain you within a flat-table db state which is not only incredibly limiting but which in practice doesn't present any advantage over a smart text editor like KEdit for example - which has come down to 99 plus VAT from 129 plus VAT, btw. Again, as soon as you have a grid and the necessary scripting know-how, and are willing to do the necessary work, I do NOT see in which way ANY db frontend would help you in ANY way, in their current state, and that's where that (otherwise awful) Access, from Microsoft, comes into play, since they say it's easy to get the necessary forms, necessary for multiple-table management* (remember: db frontends are only "good" for multiple-table-SELECTS (and when you enter all the necessary commands yourself anyway, so why not do that with a grid, too?).

* = "management" not being querying databases for data sets here, but creating new datasets and updating existing ones; as for the "management" in the sense of structure-altering commands, we've seen above that you're well advised to NOT rely on the frontend for such things either (Navicat and "Exper" mixing columns up, in different scenarios), but to write those "management" commands in sql yourself, too. Thus, if you permit the low pun, db frontends fall FLAT, in quite some ways.

ital2

  • Member
  • Joined in 2017
  • **
  • default avatar
  • Posts: 115
    • View Profile
    • Donate to Member
Re: SQL / SQLite for Beginners - Tools, Tips, Tricks, and Traps
« Reply #7 on: July 21, 2017, 06:34 AM »
SQLite frontends:

SQLabs SQLite Manager is really unbearable in Windows. Above and in the Trial thread, I related my adventures with its previous (on my XP system), I now tried to trial its current version (on my W10 system). First of all, unchanged is the trial crippling: among others, no more than 20 (or was it 25?) records, not only for search results, but also for the data as such in the grid. Then, it crashed immediately. It seems to be a Mac program, originally, and aside from their crazy trial conditions which they inforce admirably well, they don't seem to be able or willing to treat Windows users the way we are accustomed to be treated: A trial should run for its expected duration, not crash within the very first minute that is.
.
They have got the prettiest site of all of the SQLite frontends if you ask me, and their grid isn't ugly either, thus my continuous interest in that program in spite of the crippled trial (and it seems to offer a lot for just 50 bucks, according to their site's explanations), but now I'm really finished with it. Mac developers who try to make some additional sells by translating their application to Windows should probably peer less on the money and more on Windows' specificities I'd guess in general; cf. Scrivener - I don't know Scrivener but by hearsay, but what interests me here and now about it is the fact that the Windows version seems to be programmed by completely different people (who should have access to the original source code after all), and that it has been considerably lacking behind the current Mac version, and that's it's to be feared this disdain for Windows users (or call it whatever you like to call it: what about "incompetence"?) will continue to be preserved.

SQLite "Expert" comes with almost-daily updates recently, so that's what you'd call a frenzy update rate if there is any. Unfortunately, the latest output I installed from this diligent developer crashed on my system immediately, so that's what you'd call an ever-extended beta, right? (Cannot prove it since the update file went straight to my temp folder, as described above, and CCleaner takes care of those. Btw, in the lastest installments of both W10 and FF, Click&Clean doesn't seem to work correctly (or anyway) anymore, not correctly triggering CCleaner either it seems when you want to have do it at least that, so I'm currently looking out for an alternative over there, too...), so I now have to re-install some ancient version which I took the precaution to manually save some weeks ago, in order to continue to use that pretty but currently a little bit sludgy frontend for the time being; alternatively, I could check if there's even a newest-new 3-digits-after-the-dottie update? (Well, DB Browser for SQLite's rock-solid whilst being really ugly...) Gee!


And for Regex, here in db output processing or in general, see my https://www.donation...ex.php?topic=44120.0

ital2

  • Member
  • Joined in 2017
  • **
  • default avatar
  • Posts: 115
    • View Profile
    • Donate to Member
Re: SQL / SQLite for Beginners - Tools, Tips, Tricks, and Traps
« Reply #8 on: July 25, 2017, 02:46 PM »
The following was meant as an Edit to my previous post, unfortunately I got, with multiple tries, "invalid attachment". It's a 5kb .png file called SQLiteExpert.png, from c:\, so it'd be of interest to know what in such a file could be "invalid" for this forum, perhaps screenshots cannot be inserted by edit, but only into new posts, since now, doing a new post for it, it seems to work?


EDIT July 25, 2017

Of course, some days later, there is a new version of SQLite "Expert", it's quite FIVE versions later than about 5 days ago. Probably, the immediate crash will not come now. But I've said it several times, the quite pretty GUI is really bad, functionally. So I think I should share a little screenshot in order to give an impression of what I mean, this is just an example of many (or let's say, several of) such quirks. As always, just a tiny one, so that the size demonstrates, too, that my illustrations are only for conveying my point in question issues. Here, we've got a record NOT numbered 72, but with a 4-digit number, but the whole record had been more or less readable indeed, not editing purposes though, but then, I had clicked on the plus sign above the records, in order to create a new one - which you see in the middle line, with the star in front of it, and you can see that it's perfectly unscrutable - so that's how the developer thinks people should enter the data into a new record, by "in-line" editing, ha ha ha (the multiple "(n" are default nulls):

SQLiteExpert.png