Re: [Chicken-hackers] "make check" is 40 times slower than "make" (gcc 4.9)

[Peter Bex]

On Thu, Apr 30, 2015 at 03:42:08PM +0200, Peter Bex wrote:
> I've only just upgraded to Debian Jessie yesterday (so I've never noticed
> it before, running Wheezy all this time), and tested it now.  You're right,
> it's unacceptably slow.
> 
> Unfortunately, the test was introduced because there was a bug somewhere
> in the handling of manyargs, so I'd prefer not to remove or disable the
> test.  It works fine with older GCCs, and also with clang, so I would
> classify this as a (pretty serious) regression in GCC, even if this many
> arguments (2048!) to a C function is a bit on the insane side.
> 
> Can you file a bugreport with the GCC team while we figure out what to do
> with this problem?  Maybe there's an easy fix on their end, and maybe other
> projects are also having trouble with this.

An extra data point for your report: With gcc-4.8 it's still fast, so it
really is a regression between 4.8 and 4.9.  If you're using Debian, this
may be an acceptable workaround for you: "apt-get install gcc-4.8", run
"make confclean" and pass C_COMPILER=gcc-4.8 to all subsequent make
invocations.

Cheers,
Peter

signature.asc
Description: Digital signature