On 08/09/2013 07:08 PM, Patrick Walton wrote:
On 8/9/13 4:00 PM, Josh Matthews wrote:
A brief twitter observation convinced me to dig into our build system
effeciency. First off, some numbers from a -j10 build after a `make
clean`:
origin/master:
real 3m51.867s
user 3m41.425s
sys 0m10.434s
jdm/parbuild [1]:
real 2m36.697s
user 4m9.334s
sys 0m10.177s
With the changes from my parbuild branch, we build all subdirectories in
parallel, then libmsg and libutil together, libnet by itself, then
libgfx and libscript build together, followed by the main servo crate.
It's probably possible to decouple libgfx from libnet (there are
precisely two references to the Image type from libnet), but that
wouldn't actually affect the total build time in a parallel setting,
since libgfx takes 12 seconds, libscript takes 60s, the servo crate
takes 46s, and the rest are on the low end of less than 10 seconds.
libscript is obviously the long tail, and it's going to get worse as
time goes on.
These numbers seem awfully high. It's probably worth digging into what
the problem is and filing bugs against rustc. Is this with optimization
on or off?
Patrick
These are rustc invocations with -O, so presumably with optimizations.
_______________________________________________
dev-servo mailing list
[email protected]
https://lists.mozilla.org/listinfo/dev-servo