Re: [Gnu-arch-users] Re: give us a hand with arch

[Andrea Arcangeli]

On Sun, Sep 28, 2003 at 02:58:16PM +0200, Robin Farine wrote:
> >>>>> "Andrea" == Andrea Arcangeli <address@hidden> writes:
> 
> [...]
> 
>     Andrea> However again, I'm fine that taglines are available too,
>     Andrea> so I only need explicit mode to be safer (because explicit
>     Andrea> obviously it can be implemented mathematically *safer*)
>     Andrea> and more transparent than the tagline.
> 
> But again, that's wrong. With tagline and a key in your editor bound to
> 'uuidgen', you get mathematically safe embedded tags easily.

why should I bother pressing a key when add-tag will do it for me?
without even showing me garbage. I've to run add-tag anyways for the
strict commits.

And uuidgen is only probabilistically math safe, explicit can be
implemented _completely_ math safe (unlike what I originally though).

And don't tell me you would like to implement uuidgen completely math
safe, I'm sure you wouldn't like to see in every file:

        arch-tag: address@hidden

having the above in the metadata in {arch} would be fine instead.

>     Andrea> Since I need to use "explicit" for the strict commit, I
>     Andrea> would only lose the math safety guarantee by using
>     Andrea> tagline. I see only disavantages in tagline vs explicit,
>     Andrea> if you need strict commit like I do.
> 
> No, no and no. With tagline and 'untagged-source unrecognized' you
> exactly get your "strict commit". With 'tla tree-lint' you can check

yep untagged-source unrecognized is a great strict commit already
implemented and available AFIK ;)

> that your files have valid unique tags before committing. And if you
> commit without checking first, any discrepancy would abort the commit.

that wastes cpu in the regexp btw, but that's the minor issue.

The real clash I'm talking about (that sure can't happen with explicit)
will happen across two different users adding a new file at the same
time in their own branches and incidentally uuidgen returns the same id
for both despite they're two different files. Now try to make a diff of
the two branches, I'm unsure what would it happen, I tend to tink arch
should get confused and think they were the same file and it was
somewhat related.

> You only had one valid argument in favor of explicit tagging, the so
> called "file pollution". Functionally, explicit tags only have
> *disadvantages* compared to embedded tags.

Which disavantages? The only disavantage is having to run
add/move/delete-tag and friends but that's really an advantage since
it's the only way to do strict commits.

OTOH the advantages of explicit are multiple, most notably the "file
pollution" one and the math safety in the tag-id generation, but the
list is longer, see the email to Miles.

Andrea - If you prefer relying on open source software, check these links:
            rsync.kernel.org::pub/scm/linux/kernel/bkcvs/linux-2.[45]/
            http://www.cobite.com/cvsps/