Re: [PATCH] fetch2: do not try to checksum a socket

[Richard Purdie <richard.purdie@linuxfoundation.org>]

On Thu, 2012-06-21 at 15:08 +0200, Bernhard Reutner-Fischer wrote:
> On 21 June 2012 14:16, Richard Purdie
> <richard.purdie@linuxfoundation.org> wrote:
> > On Mon, 2012-06-18 at 13:28 +0200, Bernhard Reutner-Fischer wrote:
> >> computing the md5 of a socket does not work too well.
> >>
> >> Traceback (most recent call last):
> >>   File "/scratch/src/oe/bitbake/build/lib/bb/siggen.py", line 172, in SignatureGeneratorOEBasic.get_taskhash(fn='/scratch/src/oe/openembedded-core/meta/recipes-core/uclibc/uclibc_git.bb', task='do_fetch', deps=[], dataCache=<bb.cache.CacheData object at 0xd95890>):
> >>              if task in dataCache.file_checksums[fn]:
> >>     >            checksums = bb.fetch2.get_file_checksums(dataCache.file_checksums[fn][task], recipename)
> >>                  for (f,cs) in checksums:
> >>   File "/scratch/src/oe/bitbake/build/lib/bb/fetch2/__init__.py", line 627, in get_file_checksums(filelist='/scratch/src/uClibc/ /scratch/src/oe/openembedded-core/meta/recipes-core/uclibc/uclibc-git/uClibc.machine /scratch/src/oe/openembedded-core/meta/recipes-core/uclibc/uclibc-git/uClibc.distro', pn='uclibc'):
> >>                          fullpth = os.path.join(root, name)
> >>     >                    checksum = checksum_file(fullpth)
> >>                          if checksum:
> >>   File "/scratch/src/oe/bitbake/build/lib/bb/fetch2/__init__.py", line 605, in checksum_file(f='/scratch/src/uClibc/test/inet/socktest'):
> >>              try:
> >>     >            checksum = _checksum_cache.get_checksum(f)
> >>              except OSError as e:
> >>   File "/scratch/src/oe/bitbake/build/lib/bb/checksum.py", line 78, in FileChecksumCache.get_checksum(f='/scratch/src/uClibc/test/inet/socktest'):
> >>
> >>     >        hashval = bb.utils.md5_file(f)
> >>              self.cachedata_extras[0][f] = (cmtime, hashval)
> >>   File "/scratch/src/oe/bitbake/build/lib/bb/utils.py", line 362, in md5_file(filename='/scratch/src/uClibc/test/inet/socktest'):
> >>
> >>     >    for line in open(filename):
> >>              m.update(line)
> >> IOError: [Errno 6] No such device or address: '/scratch/src/uClibc/test/inet/socktest'
> >
> > Can you give some details on how to reproduce this? I'm guessing this is
> > due to some kind of proxy configuration?
> 
> I'm using an rsync of my local clone, like (The only change i have in
> an otherwise pristine oe-core):
> $ sed -e "/#/d;/^[[:space:]]*$/d" meta/recipes-core/uclibc/uclibc_git.bb
> require uclibc.inc
> DEFAULT_PREFERENCE = "9"
> PV = "0.9.33+git0"
> PR = "${INC_PR}.1"
> PROVIDES += "virtual/${TARGET_PREFIX}libc-for-gcc"
> FILESPATH = "${@base_set_filespath([ '${FILE_DIRNAME}/uclibc-git' ], d)}"
> SRC_URI = "file:///scratch/src/uClibc/;branch=master;protocol=rsync;rebaseable=1
> \
> 	file://uClibc.machine \
> 	file://uClibc.distro \
> 	"
> S = "${WORKDIR}/uClibc"

Ah, this makes more sense now. For what its worth, the "file://" url
handler will simply strip off everything including and after the ";" so
all it really sees is "file:///scratch/src/uClibc/" but this works since
it will just copy that directory whole.

It will then attempt to checksum everything in a file:// url which is
why it finds the socket and then fails. So your change does point at a
potentially valid issue. Paul might have a better idea on which level we
might want to fix this at?

> > What is worrying me a bit is that we shouldn't be calling the checksum
> > code for git repositories at all. Whilst this fix no doubt avoids the
> > problem, I'm worried there might be something deeper going on...
> 
> Could it be that the logic to deduce the fetcher is somehow broken?
> 
> Indeed, this was failing for me a couple of weeks ago with a similar SRC_URI.
> Back then i could not use file:///mydir;protocol=git (i.e. create a
> clone from a clone living on the local filesystem) so had to resort
> back to protocol=rsync.
> There was a socket lying around in that directory, a leftover from
> some testsuite run.

FWIW, the url you want is git:///mydir;protocol=file which should work.
As of the patches today, this also works as mirror syntax so you could
even add this in local.conf as a PREMIRROR and leave the recipe
pristine.

Cheers,

Richard