[PATCH] Optimize compile script on MSYS.

[Peter Rosin]

Hi!

As suggested by Ralf, there is room for improvement in the compile
script. There is no need to convert filenames when MSYS is going to
do it anyway, especially when the conversion costs a bunch of forks.

I have tested this patch with this:


cat > filenames << 'EOF'
#! /bin/sh

foo=0
while test $foo -lt $1
do
  foo=$(($foo + 1))
  args="$args $2$foo$3"
done

echo "$args"
EOF

chmod 755 filenames

time for file in `./filenames 20 / .cc`; do ./compile cl $file; done

real    0m9.781s
user    0m5.532s
sys     0m4.609s

time for file in `./filenames 20 / .cpp`; do ./compile cl $file; done

real    0m5.547s
user    0m1.737s
sys     0m1.890s

time ./compile cl `./filenames 20 / .cc`

real    0m5.218s
user    0m3.871s
sys     0m3.158s

time ./compile cl `./filenames 20 / .cpp`

real    0m0.734s
user    0m0.106s
sys     0m0.154s


The C++ files don't exist, so this includes the time for cl determining
that and outputing an error message (which I snipped).

For reference, here's a run from before the patch:


time for file in `./filenames 20 / .cc`; do ./oldcompile cl $file; done

real    0m9.875s
user    0m5.718s
sys     0m4.483s

time for file in `./filenames 20 / .cpp`; do ./oldcompile cl $file; done

real    0m9.906s
user    0m5.498s
sys     0m4.822s

time ./oldcompile cl `./filenames 20 / .cc`

real    0m5.187s
user    0m4.026s
sys     0m2.912s

time ./oldcompile cl `./filenames 20 / .cpp`

real    0m5.156s
user    0m4.044s
sys     0m2.898s


To conclude, this shaves off 4-5 seconds when compiling 20 files with
absolute file names on MSYS. The cost when the new code isn't needed
seems negligible in comparison. I get the same 4-5 seconds improvement
when I run with this fake cl script:

cat > cl << 'EOF'
#! /bin/sh
echo "$@"
EOF

That said, don't expect too much from this as using absolute file
names is not really the common case...

Cheers,
Peter

0001-Optimize-compile-script-on-MSYS.patch.txt
Description: Text document