Re: [PATCH 09/13] drop objects larger than --blob-limit if specified

Linus Torvalds [off-list ref] writes:

On Fri, 6 Apr 2007, Dana How wrote:

quoted

...
* Avoiding (e.g.) 2GB+ files when none already exist in the repository --
 either the filesystem doesn't support anything beyond the limit,
 or we don't want to use a >31b off_t with mmap.  (Perhaps
 the latter case is completely avoided by some glibc 64b trickery,
 but is that always true?)  Only the write rollback approach can address this.

I disagree violently. 

IN THEORY only write rollback can address that. But "theory" is not 
practice, and anybody who thinks that theory is even *relevant* here is 
missing the big picture.

Although I agree the "starting offset" is a very good
compromise, if we talk about THEORY, you should be able to
notice you are going to go beyond the limit inside write_one().
At that point, we know everything we need to know about all
objects that might be involved in writing that one object: if
they have been written out, what their offsets are, and how big
their in-pack representation will be.

And I agree with Nico that rollback after a failed write beyond
the boundary may not work correctly, so if we really want to do
this safely and sanely while satisfying Dana's desire to have a
hard limit, I think something like this is needed:

 - use "starting offset" to decide when to split;

 - introduce a "hard limit", perhaps optionally, to make sure
   that the result of writing out a packfile does not overstep
   that limit (i.e. the last object written below the "starting
   offset limit" might make the pack go over 700MB).

which means you would specify 600 as starting offset limit and
680 (or something like that) as the hard tail offset limit

Again, *in*practice*, for any sane situation, if you want to fit things on 
a CD-ROM, just give a limit of 600MB, and I can pretty much guarantee that 
you'll see a slop of just a percent or two for any realistic setup. And if 
it goes up to 660MB, you'll still fit on any CD.

if you really care the result fits on a CD.