From: Bil Kleb Date: 2005-09-10T23:31:32+09:00 Subject: Re: [QUIZ] NDiff (#46) James Edward Gray II wrote: > On Sep 9, 2005, at 8:26 AM, Hugh Sasse wrote: > >> It would seem useful to be produce a file of numerical differences >> as well. Writing the subsequent patch program would open another >> can of worms, but difference plots definately have their worth. > > > That's an interesting point. Maybe we can attract Bill's attention and > he'll enlighten us about how NASA would use the data... Sorry, one of my "ignore thread" filters apparently got out of hand... We do regression testing as part of our continuous integration and currently, merely use the stock diff command to compare output with a pre-existing "golden" output. This is troublesome if you switch compiler versions, machine architectures, order of operations, or other things that merely tweak the answers in the 13th decimal place or something. Our currently hack is to do some truncation of the output before write the data. So, diff only winds up comparing a certain number of decimals, but this is, of course, is not very flexible. The patch style output really isn't important. In fact, just a non-zero exit code is enough for starters. So we need ability to compare numerical fields line-by-line, ignoring any text that might be surrounding them, but make sure that for a given line, the same number of "numeric fields" exist. Later, -- Bil http://fun3d.larc.nasa.gov