Bug 3239

I run into a strange eglibc build failure due to tasks not being executed in the correct order; it happened on a fresh build of 1.3_M5.rc2, and was, I suspect, triggered initially by a corrupt eglibc tarball; the problem went away after manually removing the offending tarball and wiping out temp, but as it could not be fixed by running -c cleanall, I thought it's worth flagging it up as it suggests some underlying problem in the sstate management.

The initial failure was in do_patch, due to incomplete src tree from the corrupt tarball; not Yocto fault, obviously, but I'd expect -c cleanall eglibc to fix this, which it did not:

  bitbake -c cleanall eglibc
  bitbake eglibc

fails in the patch task because the patch task is being run *before* the fetch and unpack tasks have finished;

If I run

  bitbake -c cleansstate eglibc
  bitbake -c unpack eglibc
  bitbake -c patch -f eglibc

this succeeds; but note that the -f to the patch task is necessary, without it the patch execution is skipped, which should not happen after -c cleansstate I think. 

So now I try to run the rest of the build:

  bitbake eglibc

but now the patch task is run *again* and fails (but not as expected because the file is already patched but because the file to patch cannot be found -- it would seem the rogue patch task is being run in the wrong directory).

The above suggests I think there is a bug in maintaining the state for, at least, the patch task.