Moot: Size vs Focus

John Peacock (jpeacock@rowman.com) wrote:
>
>Barry Scott wrote:
>
>> think of is that the Office Developer SDK sold 1,000,000+ licences.
>
>difficult case. Besides, sales of SDK's doesn't translate automatically to
>actual developers (I have the Visual Developer Enterprise Suite, and I rarely
>
>people are using, so you have to make [insert platform dependent exception here]

Ok. My question wasn't related to whether a platform was important. I was
simply asking questions I didn't know the answer to - as in how important
was it really. The answer would have helped either substantiate a need or not.

As it stands, we don't know.

Secondly, I've looked into the library.

Sure, there are 27 instances of the define. To change them to another
static element is trivial.

However, that was only a stop-gap suggestion.

The real need to avoid ANY O/S specific tricks to get svn to work everywhere
would be a configuration.

The issue there is this:

The library is designed to know that the admin directory is an offshoot
of the current location. The library knows too much about the internal
construction of its own tree.

Therefore you cannot easily tell it that the admin location is somewhere
else because the code relies on knowing the name at the time at which it
runs.

Conceivably, you could get a situation where it was attempting to process
using a configured admin path, but while running a command, it accidentally
removed the ACTUAL admin path, because the wrong configuration was supplied.

Hence, it becomes an issue to be able to assign this at run time.

Secondly, you could try to allow configuration to be installed in the
main repository. Again, what happens if the repository is accessed from
somewhere else where the naming convention doesn't work. Fails again.

Lastly, you *could* attempt to export the working copy admin directory
entirely out of the current source code tree. This is workable, but
requires much thought and planning and a way to suitably query the
admin path and adjust to an acceptable alternative method of moving the
data.

So in the end, you either have a predifined workhorse to get you to the
starting gate, or you have nothing while everyone else lines up for the
starter's gun.

Had I realized all of this when I first asked the question, I probably
would have said nothing. But I guess no one had pointed out any of this
to me, so I naively figured there might be a simple explanation.

Hardcoded working copy admin directories get us to Subversion 1.0 at a
reasonable pace.

Rewiring it for Windows vs Unix is simple enough as a patch so that
windows copies can complile silently with the an adjusted filename.

Configurability which would obviate any dependence on O/S specific criteria
is beyond our immediate grasp without moving 1.0 back very very far.

If I had the computing platform to compile the windows binaries w/ the
adjustment, I would.

Perhaps someone can volunteer for this, so that we can keep moving forward
without leaving our win32 brethren in the dust.

Or perhaps we could simply use a different define that could be added to the
CFLAGS used to compile the windows binary so that it would use the alternative
working copy admin path.

No matter what, compile time is the only viable option at present.

Anything else will hold us up unduly.

I hope this clarifies things sufficiently.

I hope no one is too upset at how things are.

Time to ante up and volunteer win32 compile resources I guess, and quit
discussing this to death.

So long and thanks for all the fish.

Shamim Islam
BA BS

P.S. I'm going back to work on what I was working on before and will be
patently ignoring this discussion for the next week or so, so leave me
out of the reply loop.

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Sep 26 00:21:38 2003