Precise strdup

[sim642]

Currently strdup (and strndup) return an unknown pointer, which can make things very imprecise. Instead we should:

• Return a pointer to freshly allocated memory, the same way as malloc.
• Create a corresponding blob as its value.
• Blob size should come from the argument string constant or blob size.
• Blob contents should copy the argument blob contents or summarize the string constant.
• strndup should additionally only have maximum length n and copy as long prefix.

[michael-schwarz]

@nathanschmidt will tackle this as it is quite related to his string domain work.

[sim642]

Now that the thesis is done, will this still be worked on or not?
I don't recall where I originally noticed this, but this may be quite crucial now because silver searcher has lots of likely-spurious race warnings related to a strndup. I don't know for sure whether this would avoid them, but we might have to do this ourselves since the PLDI paper is high priority.

[nathanschmidt]

Now that the thesis is done, will this still be worked on or not?

Sorry I haven't found the time to do so yet. I'll try to do this next week, let me know if that is soon enough

[michael-schwarz]

If you can get around to it next week, I think that is timely enough; However, please don't feel obligated to do it if you don't want to, as the thesis is officially over now.

[sim642]

After some more digging into silver searcher it appears that strdup might not be the main biggest hurdle. Rather it seems proper value analysis of thread-local variables (#876) is more crucial to get rid of unknown pointers that cause imprecise type-based memory accesses.

[michael-schwarz]

Nathan has finished his thesis now and will sadly not have time to work on this, so I unassigned him.