Introduce aggregated result files

[jrosdahl]

Background

See #218.

What to do

• Use a file format as discussed/prototyped in Introduce aggregated result files #218.
• Keep result files uncompressed by default and allow for compression using zlib if compression is enabled explicitly.
• Remove support for hard links.
  • Rationale: The hard links feature has proven to be prone to errors: a) changes to files outside the cache will corrupt the cache; b) the mtime field in the file's i-node is used for different purposes by ccache and build tools like make.

[jrosdahl]

@afbjorklund is working on this in #408.

[afbjorklund]

This work is based on the original memcached code (#30) and the manifest format (ccache 3.0).

Basically we write out the length, and then we write out the contents. Nothing fancier than that...

[jrosdahl]

From #384 (comment):

I guess the --dump-result will just create files with the contents, rather than writing them on console ?

I haven't thought about it earlier, but spontaneously I would suggest a --dump-result which only describes headers and some other option (--unpack-result?) which writes the contained files to the file system.

[afbjorklund]

Initial implementation is ready, seems to be passing the tests at least.

[afbjorklund]

This implementation is still very file-centric, since it takes a cache path rather than a cache key.

e.g. ~/.ccache/1/0/066a3f1308f635a823b92939c8a3de-17195.result
instead of something like 10066a3f1308f635a823b92939c8a3de-17195

Also the implementation uses the gzFile functions, when working with the manifest and result files.

---

It's not a problem for this feature, means there are more things left to do for other features.
For instance it won't work with database backends, and won't work with LZ4 compression.

[jrosdahl]

Status: #408 and some additional work is now on master.

There are still some minor things to do, so I'm keeping this issue open for now. Thanks again @afbjorklund for the initial implementation!

[jrosdahl]

I consider this done now. I still have some related refactoring on my to do list, but that'll be done sometime later.