Re: sqlite3

[Lars Ingebrigtsen]

Eli Zaretskii <eliz@gnu.org> writes:

> The multi-test-files-busy test is sometimes failing:

Ah, right.  That's more understandable -- the one I couldn't wrap my
head around was `multi-test-files-simple' failing, because the other
Emacs should have exited and the file should be in an unambiguous state
then the first Emacs reads it.

Have the changes made `multi-test-files-simple' never fail now?

>   insert-file-contents("c:/DOCUME~1/Zaretzky/LOCALS~1/Temp/emacs-test-
>   [...]
>   Test multi-test-files-busy condition:
>       (file-missing "Opening input file" "No such file or directory" 
> "c:/DOCUME~1/Zaretzky/LOCALS~1/Temp/emacs-test-0HFBx5-multisession/files/multisession/multisession--sbar.value")
>
> This is because rename-file is not an atomic operation on MS-Windows:
> when the target file already exists, we delete it and then rename the
> new file.  So there's a small window between the delete and the rename
> that the file doesn't exist under its expected name, and the test
> fails.
>
> Does multisession.el know that a value file was already created and
> was available at some point in the past?  If so, could it perhaps test
> the file for existence before re-reading it, and if it doesn't exist,
> retry a few times, sleeping for 10 msec between retries?  (Only on
> MS-Windows, of course.)  That might solve the problem.

Yes, that sounds like a reasonable approach.  I'll have a go at it, but
I probably won't have time over the next few days, so if you want to
tinker with this instead, that's fine with me.

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no