[KDE/Mac] The meinproc4 segfault is finally REPRODUCIBLE

[Ian Wadham]

Hi René,

On 07/05/2015, at 9:47 AM, René J.V. Bertin wrote:
> On Thursday May 07 2015 Marko wrote:
>>> he verified that building only one job at a time solves this issue.
>>> What next?
> 
> Supposing this is only about meinproc4 and given how old the code apparently is, I'd not spend a lot of time on it myself. Rather, I'd try to find a suitable solution that makes sure not more than 1 meinproc4 process is active for a given project (or on the system). A lock file approach that blocks a new instance as long as an already running instance hasn't terminated should do the trick.

Good workaround.  Are you volunteering?  Unfortunately I have long
ago forgotten any details I once knew about locks, etc. on UNIX-like
systems… :-(

Concurrency is probably not the main issue, but maybe some quirk,
perhaps in XML libs, that does not like simultaneous processing of
disparate Handbook and manpage doc types.  However, it would not
hurt much to block all concurrent meinprocs because they usually run
for only a few seconds each.

> I presume you've seen backtraces showing where the crash occurs, and also know if it's always the same process (the newer or the older) process that crashes? If the crash occurs in code processing a temp file that both processes try to access, locking that file would probably even be the proper fix...
> 
> Have fun, and here's to "it" not being a Heisenbug ;)

Heisenbugs used to be my stock-in-trade, but not any more… :-(  We
used to have names for the really difficult ones.  Here's a ripping yarn
about one called the Phantom Punter, which I failed to solve…  Enjoy!
http://excda.site44.com/rimfire.html#phantom

Cheers, Ian W.