2.6.27.2[23]-libre1, 2.6.28.10-libre2, 2.6.29.[23]-libre, more news

[Alexandre Oliva]

All live at http://linux-libre.fsfla.org/pub/linux-libre/releases/
now also available in tarballs compressed with lzip.

I noticed the 2.6.28 and 2.6.29 tarballs fail deblob-check without
arguments, but pass with -B or -C.  That's not because they contain
objectionable stuff, but rather becuase of a bug in deblob-check, that
I fixed in preparation for the 2.6.27-libre1 series.  I haven't updated
the deblob-check scripts in the 2.6.28 or 2.6.29 series, though.

I expect to roll out a 2.6.29-libre1 series some time soon, for a few of
reasons aside from the bug above:

1. there's an error in the deblobbing of the WIMAX drivers, that causes
the names of the disabled non-Free firmwares to not be removed.  The
names don't even get to firmware requests, but they're still present in
the source code.  As such, I've deemed this as non-critical, but still
something I'd like to fix in an update.

2. there's something wrong with the deblobbing of the e100 driver.  I've
got an e100.ko-compatible card that does not require firmware to run,
but that is requesting firmware and failing with 2.6.29-libre.

3. I suspect there's something wrong with the deblobbing of the radeon
driver.  A video card on one of my boxes worked fine with 2.6.27-libre,
but after upgrading it to 2.6.29-libre along with a bunch of other stuff
(including the X driver) it stopped functioning.  I'm yet to look into
why, but there might be deblobbing changes involved.

If you know of any other such regressions, please let me (or, better,
the list) know.


I've been playing with a perl-based sed implementation, psed, to run the
huge sed script we generate as part of the execution of deblob-check.
After trying to generate a flex-based lexic analyzer and hitting some
practical limits (*way* too many states), I spent some time trying s2p
and psed.

It took a few adjustments to the regular expressions and to the psed
script, but once it worked, it used about 10 times less memory to
recognize, AFAICT, the same patterns.  More testing is needed, but I'm
pretty much decided to switch to a tweaked version of psed, or perhaps
to undertake a rewrite of deblob-check's main sed script in perl.  The
rewrite would be useful because the bulk of the time spent in psed is in
translating the sed script to perl, and the regular expressions in use
are regular enough (in a different sense) that translating them would be
relatively easy to do mechanically.  Generating a perl script upfront,
rather than generating a large sed script and then translating it, would
make deblob-check *very* fast again, even when using the full “-i
linux-*.*.*” pattern set.

That said, I'll probably stick with the psed-based approach for
2.6.29-libre1, so as to not delay 2.6.30-rc further.  I've got about
half-way through 2.6.30-rc3 a bit ago, but then rc4 came out, and then
rc5, and I haven't got back to it yet.  2.6.26-libre3, using the new
reject_firmware() infrastructure, is also partially done, but no ETA :-(

And then, last but not least, final artwork is needed to print T-shirts
in time for FISL (June 24-27).  Contributions would be very welcome.

Thanks,

-- 
Alexandre Oliva, freedom fighter    http://FSFLA.org/~lxoliva/
You must be the change you wish to see in the world. -- Gandhi
Be Free! -- http://FSFLA.org/   FSF Latin America board member
Free Software Evangelist      Red Hat Brazil Compiler Engineer