Thomas Gouverneur wrote:
I've tried to make a PoC for all this...
You're quick :)
Can you please change the header in the generated xref file to match the one
from the original, ie:
## PATCHDIAG TOOL CROSS-REFERENCE FILE AS OF Jan/26/12 ##
PCA is quite picky with the syntax of the first line and won't accept the file
if it doesn't match. You can put the "FOR BUNDLE xxx" part into another comment
line (they are ignored by PCA).
Even in a simple example with just one patch on the list, which pulls in a
kernel patch, I once again realized how complicated all those patch dependencies
have become, and how difficult it is to prove correctness. One thing I noticed
is that the xref file contains lines like "119254|14|Jan/01/70| | | | ||||". It
seems as if the xref creator doesn't check whether a patch has already been
included with a higher revision, and then adds these half-empty lines. When
different patches require different revisions of some other patch, only one line
with the newest required revision should be included.
I also saw a bad patch ("B") included in the xref file (120272-12). I guess the
script should look for the patch which obsoleted it and include that one instead.
The format of the lines themselves seems to be OK. Are you generating them from
your patch database or are they taken from your archive of patchdiag.xref files?
Guess that more people have to try that with real-world examples to see whether
the output makes sense and delivers a patch set which can indeed be applied.
Trying to analyze that with a handful of theoretical samples seems just too
complicated and cumbersome.
Martin.
