Is there any sane reason why we use different hashes for

(1) user display of patches and
(2) patch file names?

I had always assumed that these are the same things but a while ago I
realized that this is not so. This makes it hard to find a patch file
(for debugging).

Oh, never mind, I figured it out on my own: the 'user level' hash is
supposed to be invariant under commutation, that's the idea of this sort
of hash. Whereas the file name of a patch must be different for each
'variant' of the patch, i.e. it must change when the patch gets modified
due to commutation.

Indeed.

Since we are on this topic, I recently was thinking that it is a pity
that patches, inventories and pristine files do not have the same file
name format of just being the hash of their contents.

Patches and inventories have a prefix that indicate the size of the
uncompressed contents. I reckon this is for verification reasons, but we
could as well do without it, no?

The global cache would just be a big pool of hashed files, no matter if
they are patches, inventories or pristine. Well, even the contents of
_darcs could be unificated this way.

gh: I agree on all accounts. Have you thought about  an upgrade path /
compatibility?

I have renamed the issue and marked as wishlist.

* for simplicity sake, the cache and the contents of _darcs need to
handle the same file names

* changing the contents of _darcs means breaking backwards
compatibility, ie, older versions of darcs would not be able to read
common repositories

* however hub.darcs.net could be useful to maintain some backwards
compatibility with older darcs versions to some point

* in the light of what happened with the darcs 1 to darcs 2, my personal
opinion is that it is not worth it to allowing several incompatible
upgrades in parallel. I think we should rather wait to have a new (and
correct) darcs-3 patch kernel and update to such new repo format with
all of the changes at once.