json-c (and possibly others) memory leaks?

[50b5]

I'm pretty sure this is an issue in Android's libc but I'm surprised at how little information (including from google's git site) I've found about it.

json-c/json-c#668 and json-c/json-c#705 explain the problem and I can confirm any time I use json_tokener_parse or the test program they wrote, it leaks memory.

I wasn't sure where to post this since I could just be missing something obvious but has anyone encountered this?

[50b5]

So I did some digging.

locale_t newlocale(int category_mask, const char* locale_name, locale_t /*base*/) {
  // Are 'category_mask' and 'locale_name' valid?
  if ((category_mask & ~LC_ALL_MASK) != 0 || locale_name == nullptr) {
    errno = EINVAL;
    return nullptr;
  }

  if (!__is_supported_locale(locale_name)) {
    errno = ENOENT;
    return nullptr;
  }

  return __alloc_locale(__is_utf8_locale(locale_name) ? 4 : 1);
}

Taken from bionic master and turns out that it just ignores the base argument. So any program that tries to use newlocale as POSIX intended will leak memory. I'm bringing this up here since I don't know if json-c is the only package under termux that uses newlocale this way. Any thoughts?

Sorry for the bump btw. Just trying to figure out what to do about this.

[licy183]

Disable locale in json-c in #11220.

So any program that tries to use newlocale as POSIX intended will leak memory. I'm bringing this up here since I don't know if json-c is the only package under termux that uses newlocale this way. Any thoughts?

Yeah, I think it is difficult and almost impossible for us to find out all the problematic packages. Unless user reports these questions I think we have no way to know...

[50b5]

Thanks! Wow. Would it be possible to get bionic devs to change that? I don't see a reason why that argument is just ignored but I suppose if they built around it that way it could be a pretty major change they might not accept. I guess that doesn't even include how long it'd take to roll out...

Well I'm new to this side of GitHub but I can look through some more packages and try to help out. Thanks again!

[licy183]

Thanks! Wow. Would it be possible to get bionic devs to change that? I don't see a reason why that argument is just ignored but I suppose if they built around it that way it could be a pretty major change they might not accept. I guess that doesn't even include how long it'd take to roll out...

Well, I think this problem is not really related to bionic libc. As stated in the Linux manual page of duplocale and newlocale, Each locale object created by duplocale()/newlocale() should be deallocated using freelocale(). So I think it is an upstream issue of json-c.

Well I'm new to this side of GitHub but I can look through some more packages and try to help out. Thanks again!

Welcome to submit Pull Request or Issues if you find other packages in Termux using duplocale without freelocale!