strip: Use libelftc_string_table to manage the section string table.

[markj]

Summary

This replaces some hand-rolled routines. libelftc uses a hash table to
speed up lookups, so this ends up being way faster.

The previous implementation does suffix matching, which libelftc does
not. However, I think this is fine; in general it is unlikely that one
section's name is a suffix of another. For .strtab we'd want to do
suffix matching.

[markj]

Destroy the section string table before destroying the corresponding
ELF section descriptor.

[emaste]

The previous implementation does suffix matching, which libelftc does not. However, I think this is fine; in general it is unlikely that one section's name is a suffix of another. For .strtab we'd want to do suffix matching.

I agree this is probably fine - suffix matching can come into play with -ffunction-sections but in that case it's an entry in strtab that's a suffix of the section name.

[markj]

In D20473#443065, @emaste wrote:

The previous implementation does suffix matching, which libelftc does not. However, I think this is fine; in general it is unlikely that one section's name is a suffix of another. For .strtab we'd want to do suffix matching.

I agree this is probably fine - suffix matching can come into play with -ffunction-sections but in that case it's an entry in strtab that's a suffix of the section name.

It also would not be too hard to add it to libelftc_string_table, especially since we are allowed to break the API. Before finalizing the string table image, qsort() the entries by reverse alphabetical order and do matching on adjacent entries.

[markj]

In D20473#443087, @markj wrote:

qsort() the entries by reverse alphabetical order

I mean, sort by treating the end of the string as most significant instead of least significant.