SQLite and Mac OS X
SQLite is quite a popular library, particularly on Mac OS X where it's actually used to implement some system features.
Unfortunately, some of the filesystems on OS X (AFP, for instance) don't implement advisory locking via the POSIX function fcntl(), which leads SQLite to report that the "database is locked" whenever you try to use it.
Now, SQLite is included with Mac OS X, and Apple have been kind enough to solve this problem for the version that they provide (3.1.3). Unfortunately, that version doesn't support some of the syntax that gets generated by newer versions of SQLAlchemy, which I've been evaluating as part of a new design for my company's website. So I needed to upgrade to a newer version to make it work, and since my home area is on an AFP share, I needed it to work on that.
The upshot is that I've ported Apple's fix from 3.1.3 to the latest version of SQLite (3.3.6). Here's a patch. To apply it, download the patch and the sqlite 3.3.6 sources (1.59MB). Then, in Terminal, do something like the following:
alastair$ tar xvzf ~/Desktop/sqlite-3.3.6.tar.gz sqlite-3.3.6/ sqlite-3.3.6/art/ sqlite-3.3.6/art/tmp/ ... sqlite-3.3.6/publish.sh alastair$ cd sqlite-3.3.6 alastair$ patch -p1 < ~/Desktop/sqlite-3.3.6-osx.patch
Then build it as normal. Note that if you run make test on an AFP volume, you'll spot that one or two of the tests will fail. I think this is because of the differences in semantics between POSIX and AFP byte-range locks. If you run them from a local volume instead, they succeed as expected.
If you want the new build to match the version supplied by Apple, remember to configure it with --enable-threadsafe and --enable-tcl to turn on thread-safety and Tcl support.
Comments
When I attempted to apply the patch it ended like this:
$ cd sqlite-3.3.6
$ patch -pl patch: **** strip count l is not a number
The value displayed is a lower-case 'L', not a numeric '1'.
I proceeded to eyeball the patch file, looking for an instance where it would appear that a numeric '1' was intended, and a character 'l' appeared instead, but found nothing.
Is this a normal conclusion to the patch operation?
Also, for the absolute dweeb (that would be me), when you say "build it as normal", exactly what sequence of commands should I enter, and where do I set the --enable-threadsafe and --enable-tcl options (or does it prompt for overrides?) I assume it's just as described in the README:
mkdir bld ;# Build will occur in a sibling directory
cd bld ;# Change to the build directory
../sqlite/configure ;# Run the configure script
make ;# Run the makefile.
make install ;# (Optional) Install the build products
Correct?
Will the configure script prompt me for the overrides, or do I need to edit the configure script?
Or should I be editing the generic "Makefile.linux-gcc" makefile to specify these parms?
Posted by: davelentz![[TypeKey Profile Page]](http://www.alastairs-place.net/nav-commenters.gif)
|
July 19, 2006 10:24 AM
You've got the patch command wrong. The character after the -p is a number one (i.e. 1), not a lower-case letter 'L' (l). You should be able to see the difference in the Courier font that normally gets used for
<kbd>,<pre>and<code>tags (1 vs l).As regards building it, yes, you just follow the instructions in the README, so something like the following should work:
The two changes relative to what you wrote are that the options go on the end of the configure command, and you'll want to use sudo to do the install (it'll ask for your password so it can write to restricted parts of the filesystem).
Posted by: alastair![[TypeKey Profile Page]](http://www.alastairs-place.net/nav-commenters.gif)
|
July 19, 2006 01:14 PM