Re: toolchain/57075 (nbctfmerge looped)

[Luke Mewburn <luke%mewburn.net@localhost>]

On 23-05-26 01:35, Luke Mewburn wrote:
  | On 23-05-25 15:10, Andreas Gustafsson wrote:
  |   |  lukem%NetBSD.org@localhost wrote:
  |   |  > This may have been fixed by:
  |   |  >   https://mail-index.netbsd.org/source-changes-hg/2022/05/31/msg356307.html
  |   |  >
  |   |  > Have you seen this recently?
  |   |  
  |   |  I don't recall seeing it happen again.  But even if the bug was fixed
  |   |  by the referenced commit, it could still happen on the TNF testbeds
  |   |  since it affects the nbctfmerge tool, which is linked with the host
  |   |  libpthread and therefore may still have the bug even if the system
  |   |  being built doesn't.
  | 
  | Regarding ctfmerge: given the previous lack of error handling/reporting
  | in the concurrency / thread code in ctfmerge, and the bugs I found and
  | fixed on at least one host (macOS) that were previously not obvious
  | because of lack of error handling, I don't know if we get much value
  | keeping open older ctfmerge tickets that relate to thread issues.

To clarify, I meant:
  "older ctfmerge tickets that relate to *intermittent faults with* thread issues"

Intermittent bugs are the worst! Especially in code not using the thread
APIs as well as they should.