On Oct 23, 2004, at 5:14 AM, Leopold Toetsch wrote:
First, if you don't have it yet done, install ccache.
Thanks for the tip--seems awesome.
HOW IT WORKS...
The basic idea is to detect when you are compiling exactly the same code a 2nd time and use the pre-
viously compiled output. You detect that it is the same code by forming a hash of:
So you get really fast recompiles *except* for classes/*.c. What I'd like to have is (based on ccache's philosophy) a cache for classes/*.dump and classes/*.c files. Currently they are recreated permanently.
I started playing with this, got confused, and finally figured out why I'm confused.
In the dump case, you basically (it seems) need to parse the pmc file in order to determine the parent (chain), in order to determine the dependencies, in order to figure out what you need to put into your digest. (In the ccache case, it's the preprocessor doing this for you.) But by then, you've done most of the work I'd think (by parsing the pmc file). So at that point, you'd end up calculating a digest over a bunch of files, to avoid the Dumper overhead, and the latter may likely be faster. So this might not be a win for us, since it doesn't look like it would let us skip much.
I think this will only be worth doing if we have a faster way to determine the dependencies (basically the parents of the pmc in question). The Makefile implicitly knows this, so it might be possible for it to pass this into the script.
But I have the caching and use-the-best-digest-module logic worked out, I'm just stuck on figuring out what work it can skip. Let me know if you have any ideas, or if I'm missing something.
I guess maybe the *.{c,h} case is simpler, since there you only depend on your *.dump files (plus the script itself, etc.), though there's more work to figure out what files you're trying to generate (in the library case, etc.).
JEff
