Two source files with the same content may "collide" in direct cache about dist-clang HOT 2 CLOSED

I presented a draft of possible solution in #66: while doing direct cache search and going through a list of dependencies, we add an additional check that the input file is contained within this list. If the input file name is missing from the list, that that cache entry must have been constructed from a different source file (at least, from a different absolute/relative location as seen by compiler command line), and we avoid hitting it. This leads to the following:

All the entries with the same source file content "contend" for direct cache entry: for example, if we compile the attached code with patched version from clean checkout multiple times, then neither a/file.c nor b/file.c will hit direct cache, because after compiling a/file.c the cache entry will "point" to that file as a dependency, and the cache will miss compiling b/file.c thereafter, and vice versa: after compiling b/file.c the entry will point to that file, and if we begin another try with a/file.c, we again miss direct cache.

However, both items will be served by simple cache just fine, and if we do not "repoint" direct cache entry by compiling another file with the same content in-between, then every entry has a chance to hit direct cache too (e.g. double compilation of a/file.c alone will give direct cache hit as before).

I think this rare direct cache misses for source files with the same content are worth paying for possible false positives.

from dist-clang.

Fixed in #67.

from dist-clang.