On Fri, 2004-12-31 at 08:30 -0500, D. Richard Hipp wrote: > The issue with the third approach is deciding when to clear the > precompiled statement cache. Precompiled statements use memory > and we do not want them to hang around forever.
Why not? Programs that "generate" SQL are often-foolish. Penalize them. Perhaps that's too draconian. Is it possible for the compile command to tell anything about what's calling it (filename, line number; like perl's caller() function). If so, simply reserve one cache entry per exec per file+line.
