Re: [PATCH v5 09/35] lockfile.c: document the various states of lock_file objects
- From
Michael Haggerty <mhagger@alum.mit.edu>
- Date
- Sep 22, 2014, 15:20 UTC
- Message-ID
- <54203E4D.7030800@alum.mit.edu>
- In-Reply-To
- <20140916210317.GJ29050@google.com>
On 09/16/2014 11:03 PM, Jonathan Nieder wrote:
Show 25 quoted lines
> Michael Haggerty wrote: > >> --- a/lockfile.c >> +++ b/lockfile.c >> @@ -4,6 +4,63 @@ >> #include "cache.h" >> #include "sigchain.h" >> >> +/* >> + * File write-locks as used by Git. >> + * >> + * When a file at $FILENAME needs to be written, it is done as >> + * follows: > > This overlaps a lot with the API doc, which makes me worry a little > about it going out of date. Would improving the API doc help, or if > aspects are especially relevant here, is there a way to summarize them > more quickly to avoid some of the redundancy? > > [...] >> + * A lock_file object can be in several states: > > Would it make sense for this information to go near the definition of > 'struct lock_file'? That's where I'd start if I were looking for > information on what states a lock_file can be in.
I agree with your point about overlap. I will split the documentation into two parts with less redundancy:
* Documentation/technical/api-lockfile.txt: How to use the API.
* lockfile.{c,h}: Internal implementation details.I think the implementation details would get lost among the thousand things in cache.h. So instead, I will add a commit on top of the patch series that splits out a lockfile.h header file (which I think is a good idea anyway), and moves the internal documentation there. Sound good?
Michael
-- Michael Haggerty mhagger@alum.mit.edu