mirrors/git-annex
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages 1 Wiki Activity Actions

comment

Browse source
This commit is contained in:
Joey Hess 2019-08-15 21:44:14 -04:00
parent 7cd2120b23
commit 61bc26dc40
No known key found for this signature in database
GPG key ID: DB12DB0FF05F8F38
1 changed files with 36 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

36
doc/bugs/fresh_build_for_neurodebian__58___test_failure/comment_6_357f297d1afab22a59b2bafe4b85facd._comment Normal file
View file

@ -0,0 +1,36 @@
[[!comment format=mdwn
username="joey"
subject="""comment 6"""
date="2019-08-15T21:55:43Z"
content="""
It is indeed detecting that the file it was sending
appears to have been modified after the download started.
Probably the file has not really been modified. Instead the inode cache
for it is somehow wrong.
An instrumented run had cached:
InodeCache (InodeCachePrim 21765306 20 (MTimeHighRes 1565905801.196804558s)
InodeCache (InodeCachePrim 22158907 20 (MTimeHighRes 1565905801.196804558s))
And after the download the file it was sending had:
InodeCache (InodeCachePrim 21765305 20 (MTimeHighRes 1565905802.380791792s))
Note that the test suite moves the same file from origin,
and then moves it back, and a while after that get fails.
So at least one of the cached inodes is for the old copy of the file.
The other one is probably the work tree copy.
Still have not had luck reproducing outside the test suite with a tight loop
of move --from and --to and get.
Hypothesis: The test suite uses Annex.main for many of its runs of git-annex,
and so sqlite does not do whatever it normally does atexit. If flushing
a change to the db file gets deferred for whatever reason, a later call to
Annex.main might see old information.
If so, converting the test suite to run git-annex instead of Annex.main
would avoid the problem.
"""]]