On Tue, Aug 26, 2014 at 3:13 PM, David Woodhouse <dwmw2 at infradead.org> wrote: >> I think that's too much of a complexity, for such an issue. That could >> also very easily break in a rewrite of internal part of libopenconnect >> and would result in hard to find issues. Would it make sense to >> introduce new entry functions that will simply allocate the data they >> need? That would solve the issue, and will not require tracking each >> and every variable used internally. > But don't we already *have* entry functions in library.c for almost all > cases where this is an issue? Where the original code used to just set > variables such as vpninfo->host directly, we added functions like > openconnect_set_hostname() to library.c. I meant to add openconnect_set_hostname_const() which will simply strdup() its input and avoid messing with the caller's memory at all. > So if we just make all the functions in library.c do an appropriate > strdup() + free_func() to 'internalise' the string, isn't that 99% of > the job? There are a couple of cases where it *isn't* in library.c IIRC, > but certainly we shouldn't need to worry about future rewrites of > internal code, if we do it on the way *in*. So you suggest adding an openconnect_set_user_free(), which will be used by openconnect_set_hostname() and friends? I don't like it much, but it requires less changes and it would work too. regards, Nikos
