One more addition to dependencies list

Bruce Dubbs bruce.dubbs at gmail.com
Wed Oct 3 19:47:12 PDT 2007


Bryan Kadzban wrote:
> Bruce Dubbs wrote:
>> Chris Staub wrote:
>>> gzip <$? >$@
>>>
>>> Everywhere else when it needs to use a program in the sourcedir, it
>>> just uses the pathname..."./zdiff", "./zgrep", etc...and at the
>>> end of the testsuite it does also use "./gzip".
>> Sounds like an error in the test script to me.  We ought to fix it in
>> the book and send a patch upstream.
> 
> It might be intended.
> 
> If you're trying to test just the decompression part of zdiff (which is
> presumably shared with gunzip?), then it makes sense to me to use the
> assumed-good system gzip to build an assumed-good gzipped file.  Then
> use the just-built decompression code, and compare the result with what
> you started with.
> 
> If there was a "symmetric" bug in both the compression and decompression
> sides of gzip, then using the just-built gzip would hide it.  As long as
> the gzip file format never changes, it should be OK to use the system
> gzip binary, I think.

That's possible, but I think unlikely.  If it is the case, it should be,
at a minimum, documented in the Makefile.

In any case we need to fix it for our purposes.

> (But it may be good to bring this up upstream anyway, just in case.)

Agreed.  Since Chris found it, I think he is the most logical one to
elevate the issue unless he doesn't want to do it.  Chris?

  -- Bruce



More information about the lfs-dev mailing list