Memory-mapped files as allocation graphs

[stephenrkell]

In examples/ for some time now I've had a part-done demo allocator that understands memory-mapped ELF files, as an example of a memory-mappable file format.

This has a few interesting components.

• abstractions for positions in the allocation tree, and paths through the allocation tree, i.e. conveying strictly more information than a void*
• a generalised notion of pointers, e.g. capturing how ELF offsets are an encoding of "pointers" to other parts of the file
• a notion of "interpreter" used to give sense to these generalised pointers: they are "names" that need interpreting
• a generic traversal routine of the allocation tree or subtrees of it
• a generic printer of an allocation subtree, aspiring to print editable assembly source code that, when assembled, reproduces the subtree's memory contents byte-for-byte (but encoding "names" or "pointers" or "references" using assembly references)
• a slightly less generic printer which tries to print each part of the subtree in the most appropriate assembly notation -- e.g. if the data being printed represents instructions in the assembler's native instruction set, it should be printed as instruction syntax

This is mostly a "me thing" for the time being, but I may dump thoughts here.

[stephenrkell]

If I recall, the place where I paused the ELFy work was one where I needed to implement pointer inversion, i.e. a pre-pass to collect pointer targets, such that I could later ask "what points here?". There was a question of whether bigalloc-level support for this is useful (link to #51 also).