Dne 08. 03. 21 v 13:58 Takashi Sakamoto napsal(a):
Hi Jaroslav,
I have some concern about your commit 2cfe6addaef6[1] for alsa-lib,
It adds new library API, 'snd_ctl_elem_id_compare()', to compare a pair of IDs for control element. In the implementation, the API call returns 0 if they are the same, else negative or positive numeric value according to contents of the IDs.
I just noted that I didn't fill the function doc text correctly.
My concerns are:
- evaluation of numid field is not covered.
This is not preferable since ALSA control core implementation covers two types of comparison; numid only, and the combination iface, device, subdevice, name, and index fields. If the API is produced for general use cases, it should correctly handle the numid-only case, in my opinion.
My motivation was to allow to use this function for qsort() for example. The numid and full-field comparisons are two very different things.
- tri-state return value is semantically inconsistent
It's correct for the sorting (strcmp like).
The ID structure includes three types of values; integer, enumeration, and string. In my opinion, tri-state return value from them has different meaning each other.
True, the signess is the key, the value should be ignored.
The reason I post this message instead of posting any fix is that the fix to the API affects to alsa-utils, in which the API is used by a patch you applied a few days ago[2]. The patch also includes change to 'AM_PATH_ALSA' declaration in configure.ac with bumped-up version to '1.2.5', and it disables to rebuild alsa-utils on the latest toolchain. (alsa-lib 1.2.5 is not released yet.)
The latest alsa-lib in the git repo is already set to 1.2.5pre1, so if you upgrade alsa-lib, everything should be fine. I was a bit lazy to write a configure test and add a wrapper to alsactl to support the older alsa-lib.
Jaroslav