Re: [PATCH 1/3] Lazily open pack index files on demand

> 
> Some history of what I've been doing with git:
> First I simply had to import the repo,
> which led to split packs (this was before index v2).
> Then maintaining the repo led to the unfinished maxblobsize stuff.
> Distributing the repo included users pulling (usually) from the central 
> repo,
> which would be trivial since it was also an alternate.
> Local repacking would avoid heavy load on it.
> 
> Now I've started looking into how to push back into the
> central repo from a user's repo (not everything will be central;
> some pulling between users will occur
> otherwise I wouldn't be as interested).
> 
> It looks like the entire sequence is:
> A. git add file [compute SHA-1 & compress file into objects/xx]
> B. git commit [write some small objects locally]
> C. git push {using PROTO_LOCAL}:
> 1. read & uncompress objects
> 2. recompress objects into a pack and send through a pipe
> 3. read pack on other end of pipe and uncompress each object
> 4. compute SHA-1 for each object and compress file into objects/xx
> 
> So, after creating an object in the local working tree,
> to get it into the central repo,  we must:
> compress -> uncompress -> compress -> uncompress -> compress.
> In responsiveness this won't compare very well to Perforce,
> which has only one compress step.
> 
> The sequence above could be somewhat different currently in git.
> The user might have repacked their repo before pushing,
> but this just moves C1 and C2 back earlier in time,
> it doesn't remove the need for them.  Besides,  the blobs in
> a push are more likely to be recent and hence unpacked.
> 
> Also,  C3 and C4 might not happen if more than 100 blobs get pushed.
> But this seems very unusual; only 0.3% of commits in the history
> had 100+ new files/file contents.  If the 100 level is reduced,
> then the central repo fills up with packfiles and their index files,
> reducing performance for everybody (using the central repo as an 
> alternate).
> 
> Thus there really is 5X more compression activity going on
> compared to Perforce.  How can this be reduced?
> 
> One way is to restore the ability to write the "new" loose object format.
> Then C1, C2, and C4 disappear.  C3 must remain because we need
> to uncompress the object to compute its SHA-1;  we don't need
> to recompress since we were already given the compressed form.
> 
> And that final sentence is why I sent this email:  if the packfile
> contained the SHA-1s,  either at the beginning or before each object,
> then they wouldn't need to be recomputed at the receiving end
> and the extra decompression could be skipped as well.  This would
> make the total zlib effort the same as Perforce.
> 
> The fact that a loose object is never overwritten would still be retained.
> Is that sufficient security?  Or does the SHA-1 always need to be
> recomputed on the receiving end?  Could that be skipped just for
> specific connections and/or protocols (presumably "trusted" ones)?