Kinkie wrote:
On Fri, Dec 19, 2008 at 5:57 PM, Alex Rousskov
<[email protected]> wrote:
Kinkie,
I still think that this plan is much better:
(09:27:34 AM) rousskov: Here is an even simpler plan: write String
without ever using non-const raw(). Get it reviewed. Optimize as step 2,
if needed.
In other words, do not use any dangerous, exceptional Buffer interfaces.
Pass the non-performance review. Optimize later. Writing correct code
using safe/common Buffer interfaces should not be harder than "raw
access and explicit user-invked COW". It should be easier!
You can use const raw() where needed in String, but it should not be
needed often.
You should not need any user-invoked COW, ever (if "user-invoked" means
"explicitly initiated by user").
In the end I've ended adding a public cow() call: it's an user-invoked
optimization which doesn't hurt the semantics, and since I'm not
designing a general-purpose library I don't want to protect users
against themselves.
For the lowerCase() call I ended up not needing it, as I chose to use
public interfaces (operator[] and setAt()). Not efficient, but
efficiency can be added at a later stage if needed.
This said, I consider the framework to be reasonably complete; further
developments will be added as driven by the needs of the callers.
While waiting for review of this code by anyone willing to do it, I'll
concentrate on designing a replacement HTTP parser as an example user.
I'll try to come up with a blueprint to agree with everyone on the
general principles before trying to code a prototype.
Merry Christmas!
I've worked through the new .h and started on the .cc.
All I can see so far are documentation errors in the .cc and typos in
the .h. Patch coming later to fix those.
I've noticed there are whole sections now cut out of the old String
files. I thought that was being left untouched so the migration was not
confused by any missed old semantics on new objects?
Amos
--
Please be using
Current Stable Squid 2.7.STABLE5 or 3.0.STABLE11
Current Beta Squid 3.1.0.3
