[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Slow shell script execution on Cygwin
From: |
Robert Ögren |
Subject: |
Re: Slow shell script execution on Cygwin |
Date: |
Sat, 16 Apr 2005 02:37:17 +0200 |
User-agent: |
Mozilla Thunderbird 1.0 (Windows/20041206) |
Hi Ralf,
Ralf Wildenhues skrev:
It might be of value to retry your tests with Libtool HEAD without the
lt_ECHO='printf %s\n' setting. I had only later found out that cygwin's
ash has a builtin echo which does not interpret backslashes. I could
not find the time to test the speedup myself yet.
That, along with your improvements to the quoting stuff that slowed down
config.status, gives about 10% reduction in compile time the first time.
Making again after make clean gives about 4% reduction in time.
But still, it won't give you the speed you aim for.
I'm not unhappy, I have my ugly Perl beast that provides that speed :)
2. Is there anything (apart from cross-compiling on Linux :) ) that can
be done to increase script execution speed?
I can only speak for Libtool: however the speedup gained by cygwin
improvements may be, your numbers clearly show the usefulness for a
libtool caching mechanism. I for one would be happy to help integrating
something along your approach into Libtool proper.
That's good to hear! Are there perhaps more people on this list that see
a need for more speed?
Something to keep in mind is that the gain on non-Cygwin systems might
not be that big with a crude solution like mine. Some days ago I ran the
libtool-cache beast on my GNU/Linux setup with the adapted test script.
It needs more work, but the results so far are not that impressive (from
libtool-cache's point of view).
With Libtool 1.5.10 and bash the speedup for a full make with caching
was about 1.5, but with Libtool HEAD and ash there was virtually no
speedup at all. The caching of exe linking isn't working yet and needs
some small adjustments, but I don't expect that to give any significant
speedups. For linking larger libraries with more objects there might be
better speedup, I haven't tried.
Regards,
Robert