Tags & production questions

[Guido Ostkamp]

Hello Mark,

> IMO, you are vastly overrating the importance of all clones being able 
> to see everything at once.
>
> How big is your project?  How large are your disk drives?

to give you some numbers:

It is a large multisite project with dozens of persons having access to 
sources even from multiple countries. We have several VOBs (kind of 
Clearcase repositories). A few days ago I did an experiment and 
transferred all ClearCase versions from just the main branch of the main 
source VOB by finding all versions of all files, sorting them by checkin 
time and then replaying all checkins into a fresh Mercurial repository. 
This conversion took a whole night.

I ended up with a Mercurial repository of ~950 Megabyte size (including 
~500 MB working copy), which contained ~9200 files in ~1300 directories 
and had ~38000 changesets. A small number of files are binaries.

As I said, this was only the main branch. We also have ~15 more branches 
with main development lines most of which are still maintained where each 
branch contains numerous maintenance releases made over the years which 
are 'tagged' with labels in ClearCase.

Development mainly takes place on Sun servers running Solaris OS. In a 
professional environment, server disk space, which also has to be backed 
up at night, is very expensive - also the systems are used for a long 
time, and disks have thus not the sizes you are used to on a modern PC.

Typically, each developer has a quota of just a few Gigabytes, let say 5 
GB, which he cannot exceed.

In case you don't know Clearcase yourself, you must understand that it has 
its own filesystem which allows to define 'views' to the repositories by 
applying rulesets defined in a so called 'configspec'. You just get the 
stuff mapped in at certain directories, but there is no physical copy in 
your directory that uses up any disk space. Only compilation results like 
object files, libraries and binaries really use up storage space (at least 
in our setup where we do not use wink-in objects).

When we get a bug report for some version out in the field, we have to fix 
it for that version and port the fix to at least all newer branches 
including the mainline. This means we have to check what's in those 
versions by analyzing logs, possibly compare versions from different 
branches etc. Typically the fix is developed in the mainline first (if it 
does not already exist therein) and then ported back to maintainance 
branches.

Thus it is absolutely crucial to have everything available in one 
repository. It would be an absolute nightmare to have each of the 
maintenance releases of each branch in its own repository.

Having explained that, I should make clear that I'm just one out of many 
developers and certainly not in the position to decide which source code 
management my employer uses, so they will stay with ClearCase.

However, for my work I would like to maintain a shadow Mercurial 
repository which can help me for interim development before I check things 
back into ClearCase, this is why I try to find out whether Mercurial could 
do the job.

> You can safely clone repositories and checked out files using full 
> hardlinks with
>
> $ cp -al REPO REPOCLONE
>
> which is the fastest way to clone. However, the operation is not atomic 
> (making sure REPO is not modified during the operation is up to you) and 
> you have to make sure your editor breaks hardlinks (Emacs and most Linux 
> Kernel tools do so).

Ok, thank you for the hint. Solaris 'cp' does not have these options, but 
I think I have a GNU 'cp' available. I've also checked my 'vim' manual and 
found the 'set backupcopy=auto,breakhardlink' option which does what you 
suggest.

>> Even if there is no working copy around and hardlinks are possible, 
>> there is still a possibly large number of extra inodes used for that.
>
> Use reiserfs then.

I don't have this choice. Our Solaris systems use 'ufs' with logging.

Regards,

Guido