>le 2001/08/15 07:46, Jay Reeve à jktr@... a écrit : > >> Containers are, quite simply, handles managed by the runtime. Using them >> is generally painless, regardless of size. If, however, you are looking >> for optimal speed, there may be a small advantage one way or the other. > >It seems that containers are not exactly handles managed by the runtime, but >that andy has put an additional wrapper round the handle, probably so that >the runtime can also keep additional information, in the same way that an >edit field is a TE field at heart, but with additional FB info. i would >advise treating a container as a container, and a handle as a handle - >mixing techniques, at this stage, may result in hanging memory as the >runtime-managed stuff can't be cleared, or things become desynchronised with >the runtime info. Jonathan, I understand what you're suggesting, but I can't find the "additional" in containers as I can in dynamics or in EFs. Can you demonstrate it or show any documentation? After an admittedly brief go at containers, my conclusion is that the only difference between a container var and a handle var is that the runtime recognizes the container as a handle that it can and should manage. Has anyone found evidence of a container "header" with ancillary information, along the lines of the dynamic header that is well documented? I'm not looking to be argumentative, but if there are differences between container handles and other handles, I think we should know what they are, and their caveats and limitations should be documented. My less-than-comprehensive tests so far offer no reason to avoid treating them as equivalent. > >i also had a quick look at dynamic stuff. looks cool. a good programmer will >probably be able to do this quicker, but except in real speed humongonous >situations, this stuf will be easier. great work. The thing I find amazing about dynamics is that, in most cases, becaue of the info stored in the header, they will be faster than what most of us would come up with rolling our own! (Yay Andy!) The only ways I can see for a dedicated routine to be faster, would be by skipping a small step or two required for "generic" use, or by keeping repeatedly-used values in registers instead of in RAM--neither of which is likely to make huge differences in most real situations. 0"0 =J= a y "