Re: [Question] How to parse range-diff output

From: Junio C Hamano <gitster@pobox.com>
Date: Feb 26, 2024, 16:16 UTC
Message-ID: <xmqqmsrnfbfh.fsf@gitster.g>
In-Reply-To: <11783b8e-472d-4b75-bb48-88138daf16ae@app.fastmail.com>

"Kristoffer Haugsbakk" <code@khaugsbakk.name> writes:

Show 25 quoted lines

> On Mon, Feb 26, 2024, at 08:25, ZheNing Hu wrote:
>> Hi,
>>
>> I am currently looking to implement a service that provides a version
>> range comparison based on git range-diff. I can easily parse out
>> commit pair headers like "3: 0bf6289 ! 3: a076e88 dev5," but I am
>> unsure how to parse the details in the subsequent diff patch body.
>>
>> It is not a standard diff output where one can parse out the filename
>> from the diff header, It should be called a diff of diffs. We can see
>> various headers with file names such as "@@ File1 (new)", "## File2
>> (new) ##", or "@@ File3: function3" in different formats. This is
>> confusing. How should we correctly parse a range-diff patch, and do
>> you have any good suggestions?
>>
>> Thanks for any help.
>> --
>> ZheNing Hu
>
> Hi
>
> Note that “Output Stability” says that this output is not meant to be
> machine-readable. It’s for human consumption. It’s not textually stable.
>
> So a new version of Git might break your implementation without warning.

Good point.

In fact, those "##" things came long after the command was introduced for exactly the purpose of helping human users to locate which part of a "diff" the "diff of diff" is talking about, and the output from the command has been unstable for the exact reason.