Re: race condition in writing lilypond-book filenames

[Han-Wen Nienhuys]

On Thu, Dec 17, 2009 at 8:04 PM, Graham Percival
<address@hidden> wrote:
> On Tue, Dec 15, 2009 at 10:00:20PM +0100, Matthias Kilian wrote:
>> [sorry for mumbling to myself too much]
>>
>> On Sun, Dec 13, 2009 at 06:53:57PM +0100, Matthias Kilian wrote:
>> [...]
>> > -        out = file (path + '.ly', 'w')
>> > +        out = tempfile.NamedTemporaryFile(mode = 'w', suffix = '.lytmp',
>> > +                                          dir = directory, delete = False)
>>
>> Unfortunately, this doesn't work with python < 2.6 (because the
>> `delete' parameter is new since python-2.6).
>>
>> But if you gyys think that the approach of writing (and later
>> renaming) temporary files may help, i'll make a better diff tomorrow.
>
> I think that before anybody tries to seriously solve this, we need
> a reproducible bug.  Something that fails 5% of the time is very
> difficult to solve.  :(
>
> Can you find some abusive combination of lilypond, lilypond-book,
> make, -jX options to make, and CPU_COUNT=Y environment variables
> to pass to said lilypond-book instances, that produces a reliable
> bug?  Getting that far is easily 80% of solving this problem.

We already have a plausible explanation, and a fairly simple solution:
use flock() in ly:parse-file on the .ly file.

Overwriting files is not actually a problem, since the .ly files will
be overwritten by the same content.

-- 
Han-Wen Nienhuys - address@hidden - http://www.xs4all.nl/~hanwen