[Monotone-devel] Re: MinGW patch

[graydon hoare]

Jon Bright wrote:

> monotone.cc:189: 
> app.set_database(absolutify(tilde_expand(string(argstr))));

that one's OK; it's still dealing with the string as a UTF8, not a 
structural path. it's where it uses it where there's a problem:

> app_state.cc:35:        db.set_filename(dbname);
> app_state.cc:49:  db.set_filename(filename());

here. these are bugs. fixed.

> manifest.cc:51:  if (book_keeping_file(path()))

that's a bug. left over from an old revision; the walker never enters 
book keeping paths anymore anyways.

> manifest.cc:115:    if (!book_keeping_file(path))

also a bug; redundant code actually (see below). manifest.cc is one of 
the oldest files in the program, so it probably has some historical 
cruft in it from before I had a clearer understanding of the invariants 
I'd need.

> work.cc:44:  if (book_keeping_file(path()))
> work.cc:124:  if (book_keeping_file(path()))
> work.cc:231:  if (book_keeping_file(path()))
> These are all file_paths.  Once again, I'm unsure if as a result it uses 
> the local_path or fs::path version of the function.

the local_path variant. the fs::path variant is supposed to be static 
actually. fixed. in fact, I've just removed these checks in work.cc 
because not only is the walker never going to enter book keeping paths; 
the file_path object throws during construction and validation if it's 
inside a book keeping path. that's the whole point. it's a leftover check.

> I produced this by an amount of dark magic with awk, grep and xargs, 
> followed by manual review of the resulting list, so no guarantees that 
> it's complete, but I'm at least hopeful.

I'll see what I can do about building a cross with these changes. thanks 
for the continued investigation.

-graydon