Hi,
I'm now spending many CPU cycles on quite large machine (thanks
Dave!) to beat 20 minutes of "build" time (in an already prepared
build tree, starting with configure, bootstrap, build --all in
instsetoo_native, one language, one packaging format, prebuilt
mozilla, disabled binfilter).
I'm working on some optimalizations together with others on #dev
(football is not interesting anymore ;-).
Building our external pre-requirements from the source is an issue.
It is very ineffective :-( Caolan already reported issue #i107097#
(http://qa.openoffice.org/issues/show_bug.cgi?id=107097).
E.g. I was investigating why the build on 48 core machine spends 2
full minutes (120 seconds!) in the module curl to realize that the
authors of curl are simply "genial". They are looking for the
prototype of recvfrom with the code like this:
for tryreturntype in 'type1' 'type2' 'type3' 'type4'
for tryarg1 in 'type1' 'type2' 'type3' 'type4'
for tryarg2 in 'type1' 'type2' 'type3' 'type4'
...
compile, run the code
well, recvfrom has 6 arguments and one return type... It takes AGES
to test this. And they do so for recv, recvfrom, getnameinfo etc.
Patching this out I was able to get down to 47 seconds there which is
also very loooong on such machine...
This is simply terrible. I'm now walking over all modules to get this
sorted out a bit.
In many cases, we do not propagate -P value down to the source even
when it is supported. I'll create patches for these cases.
But there are harded cases like nss, openssl and python which are
worth deeper investigation.
Anyone interested to help with it?
--
Pavel Janík
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
