--- Begin Message ---
Subject: |
[PATCH] Remove column quoting and close sqlite db on buffer kill |
Date: |
Fri, 15 Sep 2023 11:19:28 +0000 |
Tags: patch
Hi everyone,
I had the opportunity of using sqlite-mode recently, which is really
handy for quickly inspecting the content of a database. However I
noticed two issues when using it from Windows:
- The sql query built by sqlite-mode-delete is syntactically correct,
but misinterpreted by sqlite, and eventually do nothing. The query is
of the form "REMOVE FROM table_name WHERE rowid = ? and
'column_name_1' = ? and 'column_name_2' = ? and ..." From [1], if I
understand correctly, the quoted column name in the WHERE clause are
interpreted as strings from sqlite, and as a result not a single row
is ever matched and deleted.
- The connection to the sqlite database (file) is never closed, even
when the buffer is killed. As sqlite--db is a local variable,
it's not even possible to close the connection by hand once the
buffer is killed. That means that once a database file is opened
with sqlite-mode-open-file, the file cannot be deleted unless
emacs is closed (on Windows).
Attached is the modifications that makes it work for me. I didn't have
the opportunity to test it under linux or another environment.
I did not want to touch the other parts that were already working, but
note that sqlite support a collection of pragma statements [2] that can
be used to inspect the schema of the database in a more structured why
than parsing the content of the sqlite_master table.
I hope it helps, and thanks for all the great work!
In GNU Emacs 30.0.50 (build 2, x86_64-w64-mingw32)
Repository revision: cb8f39ad6a5129fe18f2ebb0a1619f6949cd3d32
Repository branch: master
Windowing system distributor 'Microsoft Corp.', version 10.0.22621
System Description: Microsoft Windows 10 Pro (v10.0.2009.22621.2134)
Configured using:
'configure --prefix=/d/a/emacs-build/emacs-build/pkg/master-x86_64
'CFLAGS=-Ofast -fno-finite-math-only -fomit-frame-pointer'
--with-native-compilation --with-zlib --with-xpm --with-xml2
--with-tree-sitter --without-tiff --with-rsvg --with-png --with-lcms2
--with-json --with-jpeg --with-harfbuzz --with-gnutls --with-gif
--without-cairo --disable-build-details --without-dbus
--with-compress-install --with-small-ja-dic'
Footnotes:
[1] https://www.sqlite.org/lang_keywords.html
[2] https://www.sqlite.org/pragma.html
Thomas Hilke
0001-Remove-column-quoting-and-close-sqlite-db-on-buffer-.patch
Description: 0001-Remove-column-quoting-and-close-sqlite-db-on-buffer-.patch
--- End Message ---
--- Begin Message ---
Subject: |
Re: bug#65998: Sqlite-mode issue deleting records and closing database |
Date: |
Sun, 17 Sep 2023 13:05:47 +0300 |
> From: Stefan Kangas <stefankangas@gmail.com>
> Date: Sat, 16 Sep 2023 07:07:51 -0700
> Cc: 65998@debbugs.gnu.org
>
> Eli Zaretskii <eliz@gnu.org> writes:
>
> >> - The sql query built by sqlite-mode-delete is syntactically correct,
> >> but misinterpreted by sqlite, and eventually do nothing. The query
> >> is of the form "REMOVE FROM table_name WHERE rowid = ? and
> >> 'column_name_1' = ? and 'column_name_2' = ? and ..." From
> >> https://www.sqlite.org/lang_keywords.html, if I understand
> >> correctly, the quoted column names in the WHERE clause are
> >> interpreted as strings from sqlite, and as a result not a single row
> >> is ever matched and deleted.
> >
> > Thanks. I installed on the emacs-29 branch the first part of your
> > patch, which closes the DB when the buffer is killed. As for the
> > second part, I'd prefer that Lars or someone who knows SQL reviewed it
> > first, as I find it strange that Lars would code something so basic
> > which doesn't work at all.
> >
> > Could someone who knows SQL please review and chime in?
>
> Something like this in an SQL "REMOVE FROM table_name WHERE {foo}" clause
>
> 'column_name_1' = ?
>
> will check if the '?' part is equal to the string 'column_name_1', which
> is probably not what we want.
>
> Whereas this
>
> column_name_1 = ?
>
> will instead check if the '?' part is equal to the value of the column
> column_name_1 in table_name.
>
> (The "?" is just a placeholder that will be filled in with an actual
> value later.)
>
> So without having tested the code or studied it in detail, the analysis
> of the problem sounds right to me.
Thanks, so I've now installed the other part on the emacs-29 branch,
and I'm closing the bug.
> Don't we have unit tests in place for this stuff, though? Perhaps we
> should see this as an opportunity to add some...
I agree, patches are welcome.
--- End Message ---