Digest changeset anchors into purely alphanumeric strings
ClosedPublic
Actions

Authored by epriestley on Jan 22 2018, 4:22 PM.

Details

Reviewers

amckinley

Maniphest Tasks

T13045: Differential file anchors may contain characters which prevent remarkup linking properly

Commits

rP6b99aac49ddf: Digest changeset anchors into purely alphanumeric strings

Summary

Ref T13045. See that task for discussion.

This replaces digestForIndex() with a "clever" algorithm in digestForAnchor(). The new digest is the same as digestForIndex() except when the original output was "." or "_". In those cases, a replacement character is selected based on entropy accumulated by the digest function as it iterates through the string.

Test Plan

Added unit tests.

Diff Detail

Repository

rP Phabricator

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

epriestley created this revision.Jan 22 2018, 4:22 PM

Harbormaster completed remote builds in B19114: Diff 45327.Jan 22 2018, 4:22 PM

Harbormaster completed remote builds in B19114: Diff 45327.

epriestley requested review of this revision.Jan 22 2018, 4:23 PM

epriestley mentioned this in T13045: Differential file anchors may contain characters which prevent remarkup linking properly.Jan 22 2018, 4:27 PM

(See T13045 for some additional discussion about tradeoffs; I don't have a strong opinion on the complexity vs compatibility tradeoff.)

Looks good except for nitpicks.

src/infrastructure/util/PhabricatorHash.php
76–77	We've lost an entire 1/2 of a bit of entropy!! `log_2(62) * 12 = 71.45`
108	"non-alphanumeric"
111–113	For consistency, use just "digest" or "hash".

This revision is now accepted and ready to land.Jan 23 2018, 8:16 PM

epriestley added inline comments.Jan 23 2018, 8:58 PM

src/infrastructure/util/PhabricatorHash.php
76–77	Haha, yeah. I probably should have just used this approach with `digestForIndex()` in the first place since the fraction of a bit we get by including `.` and `_` isn't worth even this diff, much less whatever other weird stuff we might hit in the future. Alas, I think that ship has sailed. Almost all other cases where we use `digestForIndex()` aren't likely to ever cause issues, though, at least. On the other hand, we'd really start losing bits if we tried to deal with "0" vs "O" vs "o", "1" vs "I" vs "l" vs "i", hashes containing super offensive words like "butt", case sensitivity, and so on.

"non-alphanumeric"
Use "digest" instead of a mix of "digest" and "hash".
Clarify "index hashes", which sound like a real thing, into "digests created with digestForIndex()".

Harbormaster completed remote builds in B19147: Diff 45360.Jan 23 2018, 9:04 PM

Closed by commit rP6b99aac49ddf: Digest changeset anchors into purely alphanumeric strings (authored by epriestley). · Explain WhyJan 23 2018, 9:42 PM

This revision was automatically updated to reflect the committed changes.