if we restart the build for whatever reason, stuff gets appended as
usual, but the difference between old summary and new summary should
be more visible.
I routinely read thru logs looking for the error, and forget about which
machine it was. Having it conveniently located at the end instead of the
start saves me a few seconds.
If I had done this before, I would have fixed kill just once after privsep
and not inadvertently broken the builder stuck timeout...
problem noticed by naddy@
so write a method for that.
And don't fail silently when it doesn't work, because that causes
trouble down the line.
(e.g., your log user must be able to write into build-stats!!!)
let read_checksums normalize the path so they get recognize as one file.
-> parse it once instead of 40 times.
also, remove the old `next if md5/sha1' test, it was always just an
optimization, and becomes a pessimization now that these are gone.
1/ do subdirs on the command line first
2/ ignore errors from "preferred" dependencies (and don't register anything)
if there are actual errors beyond paths that no longer exist, we'll get them
during the next pass.
log.
Now that we can SKIPDIR, doing a first LISTING for most important dependencies
is not a trade-off for the full listing.
It allows dpb to start building most important stuff right away, so that
your machines start earlier.
Caveat: if that list of dependencies changes (ports renames), then you will
see errors because ports no longer exist.
I figure I have time to make errors in the first listing "not relevant"
before that becomes an actual issue.
- don't do anything "smart" when the queue is empty. This mostly triggers
at start, and actually hinders dpb from processing more LISTING efficiently
- lower the threshold for "next" engine computation. Engine cpu usage still
stays within a few percent, but we avoid slumps in the middle, where the
"first" rescan puts the engine out for a while (I should probably look at
a non-linear function).
maybe there's some better more sophisticated heuristics, but for now,
if you have a "nojunk" port locked, you cannot junk on that host until
you remove the lock (for good reason, at least you can "finish" that port
undisturbed).
fix for the graphics/mapnik recent burp.
