diff --using does not give Windows enough time to remove temp files

[Martitza]

This is a followup to Question #67932.

The Win32 platform (specifically) may take a long time to release diff --using temp files. This may be true for other operations on Win32, as the problem is an uncaught exception in _win32_delete_readonly().

Not all external tools are effected. For example, kdiff does not have this problem on Win32. But the very popular BeyondCompare doe shave this problem -- at least on my system.

I am offering a patch which I place into the public domain. It is primitive but works.

Martitza Mendez

[Martitza (martitzam)]

Improved patch, with improved error handing. The baseline is bzr.dev r4300.

A sleep of 250 ms (with two chances to remedy the exception, plus the initial attempt) is sufficient to produce correct behavior on the two systems I have tested. Test showed that 10 ms was insufficient and 100 ms was "almost" always sufficient. So 250 ms is a conservative value consistent with the kind of defensive behavior we want, balanced with the responsiveness of bzr. Again, this patch is activated only for Windows and only if the first attempt fails.

-M

[Martin Pool (mbp)]

Thanks for the report and patch.

[Martitza (martitzam)]

On Tue, Apr 21, 2009 at 8:25 AM, Martin Pool <email address hidden> wrote:

> Thanks for the report and patch.
>
> ** Changed in: bzr
> Importance: Undecided => Medium
>
> ** Changed in: bzr
> Status: New => In Progress
>

Thanks. I realize now that I should have marked it [PATCH] instead of
[MERGE].

I welcome all feedback on how to communicate with the bzr team.

-M

[Martitza (martitzam)]

I have verified that the fix (merged in r4400) has the intended result on win32.

This issue can be closed I think.

[Martin Pool (mbp)]

Thanks Maritza.

Feel free to change the bug status yourself.