Re: [PATCH 08/15] for-each-ref: get --pretty using format_commit_message

From: Duy Nguyen <pclouds@gmail.com>
Date: Jun 6, 2013, 00:07 UTC
Message-ID: <CACsJy8D9bLijPX7JnUEkJ4e9OU9tDN_vGa=gkkHSm3Sr34fWQQ@mail.gmail.com>
In-Reply-To: <7v61xs32n0.fsf@alter.siamese.dyndns.org>

On Thu, Jun 6, 2013 at 12:09 AM, Junio C Hamano <gitster@pobox.com> wrote:

Show 25 quoted lines

> Duy Nguyen <pclouds@gmail.com> writes:
>
>> On Wed, Jun 5, 2013 at 4:12 AM, Eric Sunshine <sunshine@sunshineco.com> wrote:
>>>> +Caveats:
>>>> +
>>>> +1. Many of the placeholders in "PRETTY FORMATS" are designed to work
>>>> +   specifically on commit objects: when non-commit objects are
>>>> +   supplied, those placeholders won't work.
>>>
>>> Should "won't work" be expanded upon? It's not clear if this means
>>> that git will outright crash, or if it will abort with an appropriate
>>> error message, or if the directive will be displayed as-is or removed
>>> from the output.
>>
>> It will be displayed as-is but that's a bit inconsistent: %(unknown)
>> prints error and aborts while %unknown simply produces %unknown. The
>> latter is how "git log --format" does it. But I think we could make
>> for-each-ref --pretty to do the former for %unknown. It'll be
>> consistent with %(unknown) and we do not need to elaborate much (it's
>> pretty obvious when it does not work).
>
> The Caveat Eric is asking about talks about "what happens to a
> %(field) that only makes sense for a commit when showing a ref
> pointing at a non-commit?", but you are answering "what happend to a
> %(invalidfield) that is not defined", aren't you?

Because %(field) is treated like %(invalidfield) in this case. I'm not saying this is the best thing to do though.

Show 6 quoted lines

> IIRC, the reason we show literal from "log --format" is to make it
> easier for the person who misspelt %placeholder to spot it in the
> output, and also make it easier for the person who use %placeholder
> meant for newer versions of Git with an older one.  It would be a
> bit unnice to die for the latter, especially if the format string is
> in a script or something.

That makes more sense for for-each-ref than log because for-each-ref is a plumbing and should support scripting. But old for-each-ref will happily reject %(newplacholder) right away. Should we take this opportunity to change this behavior in for-each-ref too?

Show 10 quoted lines

> To "log --format", all input objects are expected to be commits, so
> it does not have the "what does %(authordate) give when given a blob"
> issue.
>
> But for "for-each-ref --format", it is perfectly normal that you may
> feed a non-commit; it makes the mechanism unusable if you errored
> out %(authordate) when showing a ref that points at a tag, doesn't
> it?  Substituting an inapplicable placeholder with an empty string
> would be an easies way out, unless it learns a flexible/elaborate
> conditional formatting mechanism, I would think.

There is a blurred line here. %H (or %h) should produce an object name even for tags or blobs, but right now it produces "%H" instead. The same applies for %ad and friends, but these are clearly for commits and should probably behave like %(authordate), i.e. producing empty string. -- Duy