Make map and grep preserve their args lists. #19403
Conversation
|
I don't think leaking really is the best solution we can offer here. |
|
@Leont Given how small ARGV usually is I didn’t think it unreasonable, though obviously preserving the cleanup would be best for all. Do you have any suggestions of where to start? Some way to prevent the SAVEFREESV from nextargv from conflicting with the leave-time free of the saved $_? |
Issue Perl#19340: Previously if you mapped/grepped over an array and mutated the array within the map/grep, a segfault would happen. This fixes that by bumping the map/grep args' reference count at the start of the map/grep, then enqueueing those args for a refcount decrement at the end.
FGasper
force-pushed
the
issue_19340_argv_segfault
branch
from
9bc9e83 to
2ad6fe1
Compare
FGasper
changed the title
|
This should have been a new PR, now my first comment doesn't make sense anymore. I don't think this is going to fly either. This is one of the more difficult problems that we have, please don't try to quick-fix it. |
|
@Leont I don't mean to oversimplify, certainly. But can you give a bit more detail as to what else needs to happen? This approach doesn't incur leakage and does solve both of the reproductions in the issue thread. I don't doubt that there are bigger problems in play; I'd just like to have a better sense of those. Thank you for your time. |
|
This is a good starting point: #12315 (comment) |
|
Also see this meta ticket for an incomplete list of related tickets (the dependsOn section). |
|
@xenu: Thank you for the clarification and broader context. I would still think that fixing 30% of the cases (i.e., map and grep) is better than solving none of them. To @demerphq’s point, though, this really is a “stop hitting yourself” thing; there seems to be no reason why normal development would result in such code being written, let alone being put into production. Anyhow, I‘ll close this. |
Issue #19340: This prevents an obscure segfault.