Our toolset may have become too tight without leaving enough escape
hatch to hinder further development. "hash-object" makes minimum
sanity checks by default for a very good reason, but it means that
we cannot deliberately create broken datastreams to test against
fsck and other codepaths that are supposed to detect and report such
broken data that we may encounter in the field, perhaps created by
third-party tools.
These changes teach a new "--literally" option to "hash-object" to
allow us throw any garbage to create a broken loose object. You can
even do something horrible like
git hash-object -t bogus --literally -w --stdin </dev/null
by any garbage typename if you wanted to.
It probably needs to be accompanied by "cat-file --literally" that
does not take "-t <type>" option or does not complain even if the
contents look unreasonable. But that is for another day (hint,
hint).
The second patch is optional. I thought I may want to pass this as
a new HASH_LITERALLY bit down the callchain to index_fd(), but I
decided against it, as that will allow other codepaths to create
broken datastream, spreading this debugging aid a bit too widely for
my taste.
Junio C Hamano (3):
hash-object: reduce file-scope statics
hash-object: pass 'write_object' as a flag
hash-object: add --literally option
builtin/hash-object.c | 103 ++++++++++++++++++++++++++++++--------------------
1 file changed, 61 insertions(+), 42 deletions(-)
--
2.1.0-459-g1bc3b2b
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
