[PATCH/RFC 0/11] Building up the delta parser

Jonathan Nieder wrote:

I hear that was nigh unreadable, so here's a reroll. Less cargo-cult support for broken deltas, more readability and tests. Patches apply on top of "[PATCH 15/16] t9010 (svn-fe): Eliminate dependency on svn perl bindings". As before, the end result includes a 'test-svn-fe -d' command that can apply svndiff0-format deltas, meaning less binary garbage to worry about as you puzzle over that confusing "svnrdump dump" output in debugging sessions.

Questions?  Improvements?  Bugs?

Patch 1 is a fixup to the variable-length integer parsing code, to report unexpected EOF (i.e., declared content length too long) correctly when it occurs in the middle of such an integer.

Patch 2 is the svndiff0 interpreter in broad strokes: read window, read window, read window, .... The patch doesn't encode any knowledge about what actually goes _in_ a window aside from the header, so it will error out for nonempty windows.

Patch 3 teaches the nacent interpreter to keep the appropriate piece of the preimage in memory. This is probably earlier in the series than it ought to be, but I wanted to try out the sliding window code.

With patches 4 and 5, the interpreter learns to read the "data" and "instructions" section of a window. The effect is observable because it finds the beginning of the next window correctly.

Patch 6 is an example instruction (copyfrom_data).

Patches 7-8 introduce some sanity checks.

Patches 9 and 10 are another instruction (copyfrom_target) and another sanity check.

Patch 11 is the last instruction (copyfrom_source). That's it. You can apply deltas now!

If anything seems unclear, please don't spend time puzzling it out --- just yell at me, so the code or documentation can be cleaned up. Happy reading.