[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: 1.7.0 assert on svn_client_checkout with E235000

From: Neels J Hofmeyr <neels_at_elego.de>
Date: Fri, 04 Nov 2011 12:28:14 +0100

On 11/04/2011 10:53 AM, Barry Scott wrote:
> On 3 Nov 2011, at 11:48, C. Michael Pilato wrote:
>> On 11/02/2011 05:42 PM, Barry Scott wrote:
>>> I wish that the canonical stuff was inside the svn_client_XXX API calls and not
>>> a burdon on callers. To my mind the svn.exe API and the svn_client_XXX should
>>> accept the same strings and either operate or return an error. Avoiding the
>>> asserts from using svn_client_XXX is the lions share of the work of getting
>>> svn binding sane.
>> Two problems with such a move:
>> - libsvn_client isn't the only API called from code outside of the
>> Subversion core distribution. What does to auto-canonicalization in
>> libsvn_client do if passing the same uncanonical path to libsvn_wc will then
>> crash?
> Then the logic of the suituation would be that lib_svn_wc needed to
> change as well.
>> - canonicalizing inside the API means the API must therefore assume that
>> input paths are not canonical. That means performing the same
>> canonicalization over and over again each time the API is called. That's
>> wasteful. Better to ask the highest level of code to make its input conform
>> once (we even offer the functions to do so!) and then take advantage of
>> known-good input from then on.
> From my point of view svn_client_XXX is the highest level of code I call.
> I convert to canonical on every call of svn. Its saves nothing.

It probably saves many duplicate *_canonicalize() calls that the API would
cause. You can write a canonicalization wrapper for coding convenience, like
you did, or you can choose to handle canonicalization even earlier.

> I think it makes using the svn API unnecessarily complex.

Actually, the same argument goes for inside the API: assuming
non-canonicalized paths makes API development more complex. IMO we need to
expect valid data from API callers, for paths as for other args.

A possibility would be to provide additional API functions that are simple
canonicalization wrappers for the different real API functions -- and that
would be code bloat.


Received on 2011-11-04 12:28:51 CET

This is an archived mail posted to the Subversion Dev mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.