[Gnu-arch-users] Dirname caching for leroy's tla on cygwin

[John Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

I was doing some tests to find out why leroy's cygwin port of tla is so
slow. One thing that I found was that it doesn't remember reading the
"=dirnames" files, (these are the ones that keep track of the long name
corresponding to the number.)

I have a directory with 28 files in it. A systrace of "tla changes"
shows that it accesses 29 different "=dirnames" files. However, it
accesses them 4474 times. (strace shows a 4474 opens of a file names
"=dirnames").
Unfortunately, cygwin systrace is different than linux systrace, (at the
very least it's much more verbose. (It has 300k lines for tla changes,
versus 2.9k under linux)

Anyway, if I'm reading it correctly, each open-close cycle takes 3.4ms.
So the total time is something like 12s just for the =dirnames files. I
was thinking that it might be wise to add some sort of caching. So that
we maintain a memory map for any =dirnames files that we've read, so we
don't have to read them again.

I'm willing to contribute some time & code to adding it, but at least
right now I don't have the tools to compile tla for cygwin. I would be
willing to host the cygwin port as a arch archive
http://ct.radiology.uiowa.edu/~jfmeinel/{archives}/address@hidden
branch tla--cygwin--1.2

In the end, I want to become up-to-date with tla-1.3, but since the
patches were "proven" with tla-1.2, I figured I'd start there.

Does this sound reasonable? I will probably try to create projects for
tar,diff,diff3, etc. Just so that I can get it all out of arch instead
of downloading the source code each time. Is there an "official" archive
for these programs?

Thanks,
John
=:->


-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (Cygwin)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFA36VgJdeBCYSNAAMRAsjDAJkB1BNVBeiafnzPNaASEZZE7VJPEgCggJKg
hKzq3O5c1YNK+7CTgZm5OnU=
=oT78
-----END PGP SIGNATURE-----