Jeff King <peff@xxxxxxxx> writes: > It's inevitable that you'll have to do some data cleanup like that > anyway. Lumping together prefixes isn't flexible enough to coverall > cases. Using trailer.*.key to manually map names covers more, but again > not all (e.g., if the project used to use "foo" but switched to "bar", > but the syntax of the value fields also changed at the same time, you'd > need to normalize those, too). > > So to me it boils down to: > > - returning the data as verbatim as possible when reading existing > trailers would give the least surprise to most people > > - real data collection is going to involve a separate post-processing > step which is better done in a full programming language Yeah, I agree that these are good guidelines to use when we think about the feature. > The original %(trailers) was "dump the trailers block verbatim". But > that's not super useful for parsing individual trailers. So "only" > actually starts parsing the individual trailers. I'd argue that the > %(trailers) behavior is correct, but that %(trailers:only) should > probably be doing less (i.e., just parsing and reporting what it finds, > but not changing any names). Yes, sounds sensible.
