>>>I was surprized the HTML/XML generation takes longer (multiple times!) than measured browser time. Of course such string handling will be faster in C/C++ code, but that one was the real surprise. "Internet" = Web server and transmission times lumped together?
>>
>>Back in 2002 I had a routine which generated a very complicated table-within-table-within... (4 levels) with about 700 links in 0,7 seconds, in VFP7. Using mostly textmerge, IIRC.
>
>
>hmmm, only used textmerge to get pretty printed SQL or other strings, but never measured it against traditional assignment.
>Even if it should employ only normal variable assignment, the style should help to eliminate memory thrashing via unneccessary temp variables/steps in between.
>Do you have reason to believe it employs something else like memory mapped files?

I think it uses a series of strextract(), eval() and stuff() calls. While the last function was there since forever (I think I even saw it in Clipper 87), the other two were, I think, first internally used in textmerge as early as Fox 2.0. Then the gods saw that these were good, and made them public.

When I see some of my old code, from the nineties, where I used \ and \\ for output into a file... it worked, was blazingly fast, and the code was a mess. Once we got text-endtext into a variable, I never used those clumsy things again. And I think assignment with all chopped pieces being put together is equally messy.