[distcc] Issue with massively parallel compilation

[Victor Norman]

Markus,

I agree with you completely.  The idea of the timeouts in 2.17 is a great one. 
However, the timeout values HAVE to be configurable by the user, IMO.  For my
build, I get quite a few "Connect timeout"s, a few "Send timeout"s, and even a
couple of "Compile timeout"s.  For the last one, that means that I have some
files that take longer than 3 minutes to compile -- which is correct.  They are
BIG files.  And, so when I get a "Compile timeout", it means that I lose 3
minutes worth of compiling on the remote machine, and have to do it all over
again on the localhost...  

So, Martin, et al., please allow these values to be configurable.

Vic



--- Markus Stenberg <markus.stenberg at conformiq.com> wrote:

> I'm using latest distcc from Debian-unstable (which means 2.17) and testing
> (which means 2.16) among the machines I'm using.
> 
> My CORRECT job count is in tens. However, the distcc code allows for _5_
> seconds for both connecting remote host, and the preprocessing to
> finish. This, in my option, is completely bogus (the timeout is set in
> remote.c:204, and changed to 30 only at remote.c:221).
> 
> Obviously, when my (powersaving-enabled) laptop hits the road with say, -j
> 15, and the files involved are large C++ files with tons of includes, guess
> what I get..
> 
> Nx distcc[13045] ERROR: Connect timeout
> 
> >From my point of view, there SHOULD NOT BE timeout for actual
> preprocessing, and therefore as long as preprocessing is in progress the
> timeouts should not fire off at all. Changing timeouts to longer would be
> just fix that breaks at some point later on.
> 
> Obviously, the laptop in question is wrong compiling platform to start
> with, but that's why I'm just doing preprocessing there (and distcc speeds
> the whole process up nicely).
> 
> -Markus
> 
> -- 
> "Simplicity does not precede complexity, but follows it."
> 
> >From ACM's SIGPLAN publication, (September, 1982), Article "Epigrams
> in Programming", by Alan J. Perlis of Yale University.
> __ 
> distcc mailing list            http://distcc.samba.org/
> To unsubscribe or change options: 
> http://lists.samba.org/mailman/listinfo/distcc
> 



		
_______________________________
Do you Yahoo!?
Declare Yourself - Register online to vote today!
http://vote.yahoo.com