> On Sun, 24 Mar 2002 10:29:11 +0000, Tim Bunce wrote: > > >One of the reasons I'm (vagely) interested in a pure-perl DBI emulation > >is because it could get the DBI available for perl6 faster than > >waiting for a rewrite. Of course, it would only work for pure-perl > >drivers, and still be much slower than the real DBI, > > That depends on how fast the code, ran through Parrot, really works. > Will work. > > >but useful > >none the less. > > -- > Bart.
Are there any empirical performance data using perl 5.6.x with a pure perl DBI vs the XS version ? Having authored a couple of pure perl drivers, I'd certainly like to see a pure perl solution to simplify installation. I and some users of my drivers certainly noticed significant performance improvement in 5.6.x, and well crafted modules that adhere to the performance tuning issues described in Chapter 26 (i believe) of Prog. Perl 3rd ed. should have pretty decent performance, esp. for any DBMS connections over a network, or to DBMS that have any significant IO contention. Local access for small rowsets to non-transactional datasources (eg, MySQL, Access) will probably suffer, but they will continue to have the XS version available. (And while we're on the issue of performance...now, where did I put that array binding soapbox...kidding!) While making DBI a core module would be nice, I fear it won't be in real user's hands for several years (just check the number of issue reports on this forum that indicate they still use 5.004/5...). And it might open the floodgates for other modules to demand "core" status (tho in the case of GD and some XML modules, that might not be a bad thing!). As for Perl 6, I fear thats probably going to have even more of an acceptance problem, not because of inferior tech or performance, but because Perl may be a victim of its own success...for the same reasons sysadmins don't update from 5.005 to 5.6.0, they'll be even more reluctant to update to 6.0. In summary, I'd like to see a pure perl version, if only to see what the performance delta really is, and to be able to deliver DBI to tool- or access-limited users with less pain (and maybe less false-alarm bug reports on this forum!). BTW: My limited XS knowledge leads me to ask whether the pure perl DBI can be engineered to support the XS drivers via some autoload (yes, I realize the drivers themselves need to be built, but PPMs should address many users install issues). Anything to simplify the install process. My 2 cents, Dean Arnold
