Re: [PATCH 4/4] Add 'filter' attribute and external filter driver definition.

"Shawn O. Pearce" [off-list ref] writes:

ick.  What about something like this on top?  I moved the extra child
process for the input pipe down into the start_command routine,
where we can do something a little smarter on some systems, like
using a thread rather than a full process.  Its also a shorter
patch and uses more of the run-command API.

Well, I did not like start_command() that wanted to always
perform the full exec of something else for its inflexibility,
and this piles a specific hack on top of it...  Why not a
callback with void * pointer?

Or are you trying to make this interface as inflexible and
feature-limited as possible, perhaps to make it easier to
porting to Windows?

Junio C Hamano [off-list ref] wrote:

Well, I did not like start_command() that wanted to always
perform the full exec of something else for its inflexibility,
and this piles a specific hack on top of it...  Why not a
callback with void * pointer?

Well, that's because its always used to execute some external
program.  And some operating system designers once upon a time
thought that was the only way anyone would ever need to start a
new parallel thread of execution.  ;-)

But why do you want a callback here in start_command() given
that all you are doing is running a filter command anyway?
Is that so you could start a "thread" to handle the stdin
pipe?

-- 
Shawn.