[Gnu-arch-users] Re: Avoiding ancestor scan during get with revision library

[Aaron Bentley]

Miles Bader wrote:

> Aaron Bentley <address@hidden> writes:
> 
>>>Yeah, and let's add the compress-patch-log stuff to remove some the
>>>temptation to delete them...
>>
>>Compressing the patchlogs will remove the info I need.  (I need the 
>>"Continuation-of:" header)
>>
>>Unless you mean something other than what's previously been discussed.
> 
> 
> What I'm referring to _was_ previously discussed on this list, and
> didn't involve removing any information, merely changing its form.

Oh, sorry.  There was some discussion of replacing old patchlogs with
list of which ones were present, e.g.

patch-50...patch-125
patch-127...patch-150

I thought that was what you were talking about.

Now I see what you were talking about.  And especially given the
repetitive nature of tla patchlogs, I can imagine compression would do
wonders.

Quick testing on tla patchlogs;

tla directory: 4.4 M, tla.tar.gz: 203 k

Sheesh.

> If you're looking at the raw patch-log files instead of using tla
> commands, you'd probably need to change your code to deal with the
> situation.

No, I'm using cat-log.  As long as its behaviour is the same, that's
fine.  But it will be quite slow if it has to untar the file for each
separate log body.

Linux needs a file-compression format that supports random access to
specific files.

Aaron