Re: [Monotone-devel] OS-specific line endings

[Stephen Leake]

Stephen Leake <address@hidden> writes:

> I can invent an attr 'os_line_endings', and add a function to the
> 'attr_functions' hook to change the file line ending on workspace
> update. But that doesn't fix the bogus changes reported by 'mtn status'.
> In fact, it makes it worse; I only get the bogus changes now when I
> actually run Quartus. The same workspace contains Ada code, which I
> compile far more often. With this attr, I'll get bogus changes all the
> time.
>
> Looking thru the code for 'mtn status', it seems I need to modify
> file_io.cc 'calculate_indent' to apply the reverse line ending
> transformation before computing the file id.

Digging into this code further, it seems the most efficient approach is
to define a Botan pipe filter that does the appropriate line ending
conversion, and use that filter in file_io.cc read_data and
calculate_ident.

The inverse filter needs to be added to write_data.

Normal files would not use any filter; files with a 'filter' attribute
would use the filter named by the attribute.

That makes it much harder for users to define arbitrary filters; only
filters defined in the C++ code can be used.

Although I suppose we could define a Botan filter that calls a Lua hook.
I suspect useful filters will be in C++ anyway.

I don't understand why read_data uses a Botan pipe at all, currently;
any clues? Most other uses of Botan pipes involve some sort of hash
computation.

Another thing I don't understand is the Botan pipe cache. There is no
way to retrieve anything from the cache, so it doesn't seem to be a
storage mechanism. It seems to be a mechanism for delaying the
destruction of pipes. What's the rationale for that? botan_pipe_cache.hh
says what the cache does, but not why we need it.

-- 
-- Stephe