Re: [PATCH 05/15] ref-filter: abstract ref format into its own struct

Junio C Hamano [off-list ref] writes:

Stefan Beller [off-list ref] writes:

quoted

On Thu, Jul 13, 2017 at 8:01 AM, Jeff King [off-list ref] wrote:

quoted

 builtin/branch.c       | 14 +++++++-------
 builtin/for-each-ref.c | 22 ++++++++++++----------
 builtin/tag.c          | 30 ++++++++++++++++--------------
 builtin/verify-tag.c   | 12 ++++++------
 ref-filter.c           | 22 ++++++++++++----------
 ref-filter.h           | 22 +++++++++++++++++-----
 6 files changed, 70 insertions(+), 52 deletions(-)

The patch looks good to me. So some off-topic comments:
I reviewed this patch from bottom up, i.e. I started looking at
ref-filter.h, then  ref-filter.c and then the rest. If only you had formatted
the patches with an orderfile. ;)

As a reviewer, for this particular patchq, I actually appreciated
that ref-filter.[ch] came at the end.  That forced me to think.
...
I do want to present from Doc to header to code when I am showing my
patch to others, so this is probably a good example that illustrates
that the preferred presentation order is not just personal
preference, but is different on occasion even for the same person.

So when somebody wants to do a "from design and explanation to
provider to consumer", we would probably want "doc, *.h, *.c at the
top-level and then things inside builtin/ subdirectory" order.  Of
course, on the other hand, "I do not trust me not getting swayed by
the fact that a developer more competent than me wrote the patch"
reviewer would want to use the reverse order.

Can we actually express "top-level first and then builtin/*" order
with the diff.orderfile mechanism?  It's been a while since I last
looked at the orderfile matching (which was when I originally wrote
it) and I do not offhand know if we now allow wildmatch patterns and
the directory level anchoring "/*.c" like we do in .gitignore files,
without which it would be cumbersome to make ref-filter.c listed
before builtin/branch.c in a generic way.

On Thu, Jul 13, 2017 at 2:32 PM, Junio C Hamano [off-list ref] wrote:

Junio C Hamano [off-list ref] writes:

quoted

Stefan Beller [off-list ref] writes:

quoted

On Thu, Jul 13, 2017 at 8:01 AM, Jeff King [off-list ref] wrote:

quoted

 builtin/branch.c       | 14 +++++++-------
 builtin/for-each-ref.c | 22 ++++++++++++----------
 builtin/tag.c          | 30 ++++++++++++++++--------------
 builtin/verify-tag.c   | 12 ++++++------
 ref-filter.c           | 22 ++++++++++++----------
 ref-filter.h           | 22 +++++++++++++++++-----
 6 files changed, 70 insertions(+), 52 deletions(-)

The patch looks good to me. So some off-topic comments:
I reviewed this patch from bottom up, i.e. I started looking at
ref-filter.h, then  ref-filter.c and then the rest. If only you had formatted
the patches with an orderfile. ;)

As a reviewer, for this particular patchq, I actually appreciated
that ref-filter.[ch] came at the end.  That forced me to think.
...
I do want to present from Doc to header to code when I am showing my
patch to others, so this is probably a good example that illustrates
that the preferred presentation order is not just personal
preference, but is different on occasion even for the same person.

So when somebody wants to do a "from design and explanation to
provider to consumer", we would probably want "doc, *.h, *.c at the
top-level and then things inside builtin/ subdirectory" order.  Of
course, on the other hand, "I do not trust me not getting swayed by
the fact that a developer more competent than me wrote the patch"
reviewer would want to use the reverse order.

I do not understand what you say here? Are you saying:
"I can be tricked easier when the order is top-down,
specifically when the more competent developer tries to?"

Can we actually express "top-level first and then builtin/*" order
with the diff.orderfile mechanism?

By reading the code, I think we snap to the first match. And matching
is done via the wildmatch.{c,h}, that claims it has special-case '/' matching,
and '**' **  work differently than '*',

without which it would be cumbersome to make ref-filter.c listed
before builtin/branch.c in a generic way.

Indeed.