At 2025-01-28T21:07:51-0600, Dave Kemper wrote: > On Mon, Jan 27, 2025 at 6:27 AM G. Branden Robinson > <g.branden.robin...@gmail.com> wrote: > > Dave Kemper's done some solid QA work that should avoid > > embarrassment in groff 1.24.0.rc1. > > Thanks! But so that the record's not overstated: > > It's great that groff has hundreds of regression tests now.
Well, 235. Of which the latest verifies that `hys` behaves sanely in the scenario you gave it. > But a problem like the .hys request failing to work altogether > (http://savannah.gnu.org/bugs/?66723) reveals that there are still > gaps in that coverage (though .hys itself is one less gap now). Oh, sure. Heaps. With the `char` request family issues now on my plate, I expect we'll be seeing one or more new tests covering those, too. Retrofitting test coverage is seldom a giant leap. It will take a long time before we have coverage we can plausibly claim is "comprehensive", and that's considering only the formatter. Ingo's been a bit dismissive of our efforts to date because they are so scattershot, but I defend them on the basis that (1) coverage now is way better than in the groff 1.22.4 release and (2) the existing tests regularly prevent me from committing more breaking changes than I do. In my view, point 2 is worth the trouble all by itself. Also, the process of writing tests does another thing it's supposed to, according to storied principles of software engineering. It makes me think over what I'm trying to do, what I've actually done, and compels me to learn the system under test better. You guys don't see the stuff I throw away or put on a shelf, often never to return to it. The Git stash of my groff working copy has 414 items in it, many of which I'm sure are obsolete. > A focused QA effort would attack those gaps. That's not something I > have the time to take on. I'm merely reporting issues I'm finding in > my normal usage of groff. *My* normal usage may include underused > requests like .hys, but there are huge swaths of groff's capabilities > that I never touch, so my test coverage is nowhere near comprehensive. I've worked with people who wrote machine-verified formal proofs. It's a specialized discipline. Five or six years ago, I foolishly though it was a growth area, the next direction software engineering might take. Nope. AI/LLMs have taken off like a rocket. Maybe because it's easier to conceal the underlying mathematics and consequently to write bullshit business plans and pitches to venture capitalists about. Formal methods still wears its math more or less on its sleeve and that makes Big Money people yawn. At any rate, don't undersell the service you're performing. You doing more QA work on groff than anyone besides Deri. And Deri's talent is for finding mortifyingly hard problems that I can see taking me weeks, months, even years to fix. (They're not _all_ like that, but even an occasional "term project" is daunting.) More often than not, you throw me ones I can actually resolve in a reasonable time frame. I get my dopamine hit and can keep going. Regards, Branden
signature.asc
Description: PGP signature
