* fixup: dedup tr_rpc_address with tr_address
tr_rpc_address was duplicating a lot of work that tr_address was doing a
lot better. Fall back to using tr_address for ipv4/ipv6 connections and
make a tr_unix_addr for handling unix sockets.
* remove unnecessary functions
`tr_parseNum<>()` is implemented with `fast_float::from_chars()`, and
the latter is documented as "expecting a locale-indepent format
equivalent to what is used by std::strtod in the default ("C") locale".
To accomodate locale independent number parsing, switch back to
locale-independent number formatting in `tr_truncd()` by both removing a
`L` format specifier from `fmt::format_to_n()` call and using a fixed
"." decimal separator when truncating the value.
* Expose files' begin and end pieces via RPC
This adds two arguments, `beginPiece` and `endPiece`, for each of the entries
in the `files` array of the RPC's `torrent-`get` method.
The point is to allow RPC clients to display a file's completion progress as
piece-based in addition to byte-based.
* Use user-preferred locale
Previous fix adding `L` format specifier was correct but only fixed half
of the problem, as C++ locale is set up to be "C" by default. GTK client
used to call `setlocale(LC_ALL, "")` to set up user-preferred locale
which only affected C functions and `std::locale` (used by libfmt) was
unaware of those changes.
Apply the fix to all the binaries since they're all doing some sort of
output to the user and calling libtransmission helpers, as well as using
libfmt directly.
* Improve libtransmission's json-test
Set the locale C++ way to avoid any sort of inconsistencies, and also
restore it to the old one once finished testing.
* Improve transmission-show test runner script
Quote outputs to avoid CMake error about `message()` being called with
no arguments. Capture stderr to the same output file. Fallback to `git
diff` if `diff` wasn't found. A few other minor changes.
Write non-piece peer messages directly to the peer's outbuf instead of
waiting for a pulse() message to refill it. This can help with latency
sending messages out.
Change the semantics of `tr_peerIo::get_write_buffer_space()`: this is
now interpreted as the preferred minimum size, rather than the maximum.
It's OK to enqueue an outgoing piece message as long as there's _some_
space left, even if the message is larger than that space.
Build peer messages with template fold expressions. This lets us move
all the message-building to a single function and add some sanity checks
to the outgoing messages.
LaserEyess