On Tue, Mar 26, 2019 at 10:26 PM Lassi Kortela <xxxxxx@lassi.io> wrote:
> I would still wait for Ciprian's opinion on the HTML structure.

I got stalled due to other "stuff".  I don't think I'm able to resume
experimenting, until perhaps mid April.  (Although I've followed the
discussions.)

Anyway, I would say it would take me a couple of full "work-days" to
finalize my initial experiment.  (As said my main focus is on
structuring the HTML, without changing the actual text or semantic, to
better aid in "targeted" documentation extraction, which includes
indexing and back-referencing, as a first concern, and if possible
signature extraction as secondary by-product of the main task.)

> My crude
> estimate is that 3 people could do all the edits in a month by marking
> up on average 2 SRFIs per day each (about 30-60 minutes).

I would say that is a major understatement of the required effort...

Even if we assume that we just add HTML classes (without touching any
other HTML restructuring), it won't just take 60 minutes...  It might
take that long to actually "add" the classes, but it will take much
longer than that double-checking (i.e. `git diff --color-words`) to
make sure we didn't touch anything else, double-checking by hand that
our automatic tools didn't extract "too few" or "too many"
definitions, etc.

My best-case scenario, for a mid-sized SRFI (say half of SRFI-1),
would be 1-2 days (4 hours per day), to make sure we get quality work
out of that.  (And this doesn't take into account the review /
peer-review, once the initial patch is done.)

However if we also embark on HTML restructuring, I would say the
effort required will double or even tripple...

> That work is
> substantial enough that I'd like to settle the final form of the HTML
> structure we want before starting the mass editing.

Agree.

On the other related topic (that started this thread) I would strongly
agree with Arthur on minimizing the number of moving parts that result
of this endeavor.  (I.e. servers, Heroku dyno's, on-line services,
etc.)  I am strongly for a static-file approach (that can easily be
served off Netlify with CloudFlare for caching), which will keep
thinks simple and secure.

I've read Lassi's GitHub webhooks / Travis / Heroku (with database)
proposal, and although it is "current" (in terms of best practices), I
would say it is an extreme overkill for SRFI's, especially given how
many moving parts there are, on how many external services we depend,
and on the overall complexity.  (I've "heard somewhere" that Travis
was just bought, and that they are "keeping it live" for the moment,
but the future is uncertain...)

Thus my suggested approach (as Arthur also proposed) is to keep things
semi-automatic, with human-in-the-loop.

Ciprian.