Paul Licameli
2017-01-23 13:16:20 UTC
I did git push -f to my origin. There are now three commits in the
branch. New branch head is:
https://github.com/Paul-Licameli/audacity/commit/4059b2ae48b57cd075d9cd226cb591b939e5fc6c
The head includes the fork-and-crash fix, but could it be that we don't
need it, and the changes for cleanup of temporary files are enough?
To test whether that is so, you can checkout one or two commits before the
head. Using caret notation, you can name those as
4059b2ae^ and 4059b2ae^^
Two commits before head, you get just the change to the folder cleanup that
I wrote yesterday.
After seeing Gale's log messages, which suggest that removal of folders
does not succeed if done too soon after removal of the contents of the
folder, I added the other commit one before head. This inserts code to
delay between removing files and removing folders. As submitted, it simply
sleeps by a number that you can tune. But if you change #if 1 to #if 0,
then instead it tests in a loop whether it can detect completion of removal
of the files, but not exceeding the time limit.
So I suggest this testing:
1. See if 4059b2ae^^ is sufficient fo fix the problem of closing without
saving
2. If not, see if 4059b2ae^ does it, with a long enough sleep that is
not unreasonable
3. If #2 succeeds, verify that it also works when changing #if 1 to #if
0, but that the delays are now variable and shorter
4. If #1, #2, or #3 is enough to fix closing without saving, determine
whether it is enough for all symptoms of bug1567.
Also check that the log file no longer routinely contains (as master does,
even on my computer) error messages related to the temporary files.
PRL
branch. New branch head is:
https://github.com/Paul-Licameli/audacity/commit/4059b2ae48b57cd075d9cd226cb591b939e5fc6c
The head includes the fork-and-crash fix, but could it be that we don't
need it, and the changes for cleanup of temporary files are enough?
To test whether that is so, you can checkout one or two commits before the
head. Using caret notation, you can name those as
4059b2ae^ and 4059b2ae^^
Two commits before head, you get just the change to the folder cleanup that
I wrote yesterday.
After seeing Gale's log messages, which suggest that removal of folders
does not succeed if done too soon after removal of the contents of the
folder, I added the other commit one before head. This inserts code to
delay between removing files and removing folders. As submitted, it simply
sleeps by a number that you can tune. But if you change #if 1 to #if 0,
then instead it tests in a loop whether it can detect completion of removal
of the files, but not exceeding the time limit.
So I suggest this testing:
1. See if 4059b2ae^^ is sufficient fo fix the problem of closing without
saving
2. If not, see if 4059b2ae^ does it, with a long enough sleep that is
not unreasonable
3. If #2 succeeds, verify that it also works when changing #if 1 to #if
0, but that the delays are now variable and shorter
4. If #1, #2, or #3 is enough to fix closing without saving, determine
whether it is enough for all symptoms of bug1567.
Also check that the log file no longer routinely contains (as master does,
even on my computer) error messages related to the temporary files.
PRL