Version script glob support

[marxin]

Based on the discussion with David, I tested the bevy with dynamic linking (500K symbols in a version script file), and I decided to keep the exact match implementation. The only change is that glob::Pattern is used for a general glob pattern; having that, the link time is equal for bevy. For a reasonably sized shared library that uses a couple of general globs, the speed should be good!

Plus the PR support if a space(s) is present after a symbol name.

[davidlattimore]

Thanks! Definitely good to have support for more patterns. The utf-8 string thing does make me a bit uneasy, but I guess we can go with it and potentially change it to something byte-based later.

Out of curiosity, what kinds of patterns have you encountered that weren't something*? Is it just patterns that have * somewhere in the middle, or have you found use of ? or character sets?

[marxin]

Out of curiosity, what kinds of patterns have you encountered that weren't something*? Is it just patterns that have * somewhere in the middle, or have you found use of ? or character sets?

One example could be alsa-lib, where one of the versions is defined as:

$ wc -l ./work/alsa-lib-1.2.14-abi_x86_64.amd64/src/Versions
222 ./work/alsa-lib-1.2.14-abi_x86_64.amd64/src/Versions
...
$ grep '\*' ./work/alsa-lib-1.2.14-abi_x86_64.amd64/src/Versions
    snd_*;
    _snd_*_open;
    _snd_*_dlsym_*;
    _snd_*_poll_descriptor;
    _snd_pcm_hook_*;
    __snd_pcm_hw_params_*;
    __snd_pcm_sw_params_*;
    __snd_*_dlsym_*;
    *;
    _snd_safe_strto*;
    snd_ump_*;
    snd_seq_ump_*;
    snd_ump_endpoint_info_set_*;
    snd_ump_block_info_set_*;

[marxin]

The utf-8 string thing does make me a bit uneasy, but I guess we can go with it and potentially change it to something byte-based later.

Yep, my expectation is the glob patterns are not so common, and if so, then the total number of symbols is reasonably small.

[marxin]

Well, in the context of #952, I believe, the symbol identifiers should be valid UTF-8 strings :)