See:
http://www.gnuvola.org/wip/ (libredwg)
The v2 series aims to enable (optionally configured) runtime trace
control. It is inspired by a comment on the LibreDWG wiki.
Details:
f7a9d0a [build] Move config.h into src/.
It's nice to have all the source components together in one
directory. Also, this allows Automake to avoid generating "-I..".
f9b1a87 [build] Make every .c file #include "config.h".
This propagates the information found by the configure script to the
source files. Our immediate need is to pass the fact that the option
"--enable-trace" was specified. However, looking ahead, there may be
other kinds of information that will be useful, such as endianness,
alloca availability, word size, etc, and gnulib is a nice way to add
support for such things, so we make sure this change is forward
compatible with gnulib.
54455cc [build] Add preliminary "configure --enable-trace" support.
This uses AC_ARG_ENABLE, defaulting to "no".
c3e29a8 [build r] Add runtime tracing support for read (decode) path.
32bf5b3 [build w] Add runtime tracing support for write (encode) path.
These make the respective file-static `loglevel' variables in
src/{de,en}code.c subject to a lookup of the env var LIBREDWG_TRACE.
The lookup is done only once. If there is no tracing, the default
is now hardcoded as zero (no output).
fb534d3 Refill README; nfc.
2a91cbc [admin] Mention --enable-trace in README; nfc.
This documents the configure script feature. It includes a blurb:
"NB: This is an experimental feature that is not in its final form."
That's because i think "levels" of tracing is not the best; i
personally would prefer to see a bitset approach: The library-global
(as opposed to bifurcated file-static) variable `trace' would hold a
set of independent flags that enable tracing for various aspects of
the operation. For example: encoding, decoding, top-level objects,
strings, etc.
Another idea is to support a file descriptor for trace output.
Together, these features would be very helpful for "make check".
WDYT?
