StoneCypher / jssm / ba701ffc08413007a51680aac28ac1f4c1667fde / 6

100%

master: 100%

Pull #973

github

feat(charts): measure a website by its Pages deploys

fsl.tools ships a deep site, so neither the landing page nor the repo
was the right measure -- one ignores most of what exists, the other
counts source nobody downloads. The served directory is the deliverable,
and GitHub already records every time it changed.

Adds a `pages` version strategy to collect_tagged_repo_sizes beside the
existing `tags` one:

- versions are successful Pages BUILDS. A website has no tags worth
  speaking of (fsl.tools has exactly one across eight years), but a
  build is the honest analogue of a publish: the moment the thing
  visitors reach actually changed.
- errored builds are dropped -- a failed deploy never served bytes, and
  counting it would claim visitors saw something they did not. 4 of 35.
- repeat builds of one commit collapse to the earliest; a rebuild of
  identical content is not a new version of the site.
- `subdir` names the Pages source path, and the prefix is stripped, so
  the measure is the served site and the paths read as site paths rather
  than repo paths.
- a build whose source path did not exist yet is skipped rather than
  recorded as zero, which would draw a false trough.
- version keys are `YYYY-MM-DD.<sha7>`: readable in a tooltip, unique,
  chronologically sane, and honest that a site has no version numbers.

What it found, across 31 deploys:

  2018-11-18     1-3 files              scaffolding, CNAME thrash
  2018-11-19    1437 files   12.23 MB   the generated site lands
  2019-05-07    1444 files   16.67 MB   url viewer
  2021 -> 2022  1444 files   16.67 MB   flat for three years
  2026-06-04      63 files    1.34 MB   the modern rebuild

16.67 MB -> 1.34 MB, 1444 files -> 64. The same eviction story as the
npm tarball, on a surface nothing was measuring.

Chart: 14 -> 15 packages, 787 -> 818 versions, 11 -> 12 flow columns.

Note the interned-path table roughly doubled (1586 -> 3089) and the page
went 614 KB -> 951 KB: the 2018-2022 site... (continued)

3869 of 3869 branches covered (100.0%)

Branch coverage included in aggregate %.

6363 of 6363 relevant lines covered (100.0%)

5562.88 hits per line