https://bugzilla.redhat.com/show_bug.cgi?id=1283764
--- Comment #24 from Tom "spot" Callaway <spotrh(a)gmail.com> ---
Be grateful you are not a perl programmer. :)
It seems the values can be initialized but NULL and trigger this, but I cannot
make it happen locally. The comment block directly above the line of code says:
# If the file was changed the same second as it was last read,
# we only reopen it if it's length has changed. The alternative is that
# sometimes, files would be reopened needlessly, and with reset_tail
# set to -1, we would see the whole file again.
# Of course, if the file was removed the same second as when it was
# last read, and replaced (within that second) with a file of equal
# length, we're out of luck. I don't see how to fix this.
Which leads me to believe their algorithm is totally wrong.
The challenge in trying to debug this is that I simply cannot reproduce it. I
put in code that I thought would fix the issue, but you reported that it did
not. Everything seems to be initialized and the entire existing test suite
passes. I had hoped upstream would be able to provide insight (5 years ago),
but they have clearly abandoned the effort.
Given all of that, I am hesitant to make additional changes to the algorithm
logic without a reproducer. Are you still actively using a setup that could
reproduce this?
--
You are receiving this mail because:
You are on the CC list for the bug.