Week 2: What's the Difference?

Hi everyone,

I started 2 weeks ago by studying Olga’s work, but it didn’t help me much. I dug into pretty.c and ref-filter.c, still wasn’t able to find the right approach.

ref-filter is very well coded and is simple to understand. It simply uses atoms and there is parser function for almost every atom.

Here is the whole working of git for-each-ref in few lines:

filter_refs(...);
verify_ref_format(&format);
ref_array_sort()
for (i = 0; i < maxcount; i++)
 show_ref_array_item(array.items[i], &format);
ref_array_clear(&array);

Its recipe is pretty straight.

1. call ref-filter API
2. verify given format
3. sort ref_array, if sorting provided
4. fill the strbuf, based on given format
5. print refs
6. clear memory

done!

On the other hand, working of git log is very dispersed. e.g

• initialization and other stuff is happening in log.c. [cmd_log()]
• correctness verification of commit format is happening in pretty.c [get_commit_format()]
• printing is done in log-tree.c [show_log()]

The stupidest thing I tried is re-writing cmd_log() from scratch in for-each-ref’s way. I successfully implemented the “%h” using ref-filters logic. Here is why it’s bad:

• I can’t implement every single functionality of git log
• Backward-Compatibility
• Other features like format patch is also using pretty.c logic. So I cannot get rid of pretty formatting logic.
• Getting rid of duplicate code is the aim of my project. I cant see that happening with this approach.

What’s next

I’m almost familiar with pretty.c, log.c, for-each-ref.c, and ref-filter.c. I will spend time understanding log-tree.c[specially show_log() and show_decoration()] and will document log-tree.h.

Hopefully next week I come up with some good approach.