six.moves needs http.client and urllib to exist first

• Type: feature (library) — Track B.
• Filed: 2026-08-18 by frank3-fc from [[feature-b-module-shims-for-the-html5lib-corpus]], where it was measured and deliberately not written.

Why it is not a shim you can just write

six.moves exists to give one name to a module that moved between Python 2 and 3. Every member is a re-export, so mimic_six_moves.py cannot contain anything — it can only forward to modules this build does not have. lib/rtl/mimic_six.py already says so in its DELIBERATELY ABSENT note; this ticket is that note ranked as work.

What the corpus actually imports

html5lib/_inputstream.py     from six.moves import http_client, urllib
html5lib/filters/sanitizer.py  from six.moves import urllib_parse as urlparse

plus two files that inherit the wall transitively (html5lib/__init__.py, html5parser.py), which is why the row counts 4.

The two halves, and they are not equally hard

• urllib_parse — urlparse, urljoin, quote, unquote, urlencode. Pure string manipulation against a published grammar (RFC 3986): writable exactly, testable as a differential against CPython, no backend. This is the tractable half and it unblocks sanitizer.py on its own.
• http_client — needs a real HTTP client, which is [[feature-b-mimic-urllib-request-over-the-rtl-http-stack]]. _inputstream.py imports it to fetch a document by URL.

Doing the first half alone is worth it and does not commit to the second. Note _inputstream.py imports both in one statement, so it stays blocked until http_client exists regardless.

Caveat carried from the parent ticket

_inputstream.py and html5parser.py currently stop on six_moves FIRST and have CodecInfo behind them. Do not read this row's 4 files as 4 files unblocked; measure with tools/nilpy_ladder.py after, as the parent did.

Status 2026-08-18 (frank3-fc): the tractable half is blocked too — parked

Dispatched to do the urllib_parse half alone, on the reasoning in this ticket that it is pure string work and unblocks sanitizer.py by itself. Measured before writing it, against pinned v349 (596799fd9c6e), and the reasoning does not survive contact:

sanitizer.py's import spelling does not work, whatever the shim contains.

from six.moves import urllib_parse as urlparse    # line 15
uri = urlparse.urlparse(val_unescaped)            # line 841

from M import <module> as alias followed by alias.f() is undefined variable (f) — [[bug-n-from-import-with-an-as-rename-loses-what-it-renames]] (N, p75). Without the rename it works and a top-level import M as alias works. (This ticket first said "renaming a FUNCTION works" — wrong, and corrected in the bug: a renamed function loses its signature, so a zero-argument call SEGFAULTS and an omitted default is dropped. The one-argument case I first measured is the shape that happens to survive.) And it is NOT the shim mapping: two plain modules reproduce it identically.

So mimic_urllib_parse.py would today unblock zero files — the same measured test this campaign applies everywhere, and the same answer that got xml.dom refused three times. Writing it now would produce a correct module sitting behind a wall, and would make the ladder row disappear into a different error without a file moving.

Also learned while measuring, and worth keeping

A shim CAN re-export another module; the constraint is narrow and specific:

So when this unparks, mimic_six_moves.py must re-export by the literal mimic_urllib_parse filename, and that is a workaround to register in devdocs/dev/track-b-workarounds.md against [[bug-n-from-a-shim-import-a-class-loses-its-class-level-attributes]]'s family — the platonic spelling is import urllib.parse as urllib_parse.

What unparks it

The p70 above. Then: write mimic_urllib_parse.py (urlparse/urlsplit/urljoin/ quote/unquote/urlencode, differential-tested against CPython by value, the way the five shims in [[feature-b-module-shims-for-the-html5lib-corpus]] are), add mimic_six_moves.py re-exporting it, re-run the ladder, and report past-vs-onto. The http_client half stays out of scope — that is [[feature-b-mimic-urllib-request-over-the-rtl-http-stack]].

LANDED 2026-08-18 (frank3-fc) — the urllib_parse half, on pinned v351

The blocker ([[bug-n-from-import-with-an-as-rename-loses-what-it-renames]]) was fixed and reached Track B in v351, so the park is over. mimic_urllib_parse.py and mimic_six_moves.py are in, make lib-test green.

The park was the right call and this is the evidence: the shim's CONTENTS were never the blocker. Written now, in the same shape it would have had two hours ago, it moves five files. Written then, it would have moved none.

Verified by value, through the corpus's own spelling

test/lib_mimic_six_moves.npy reaches urllib.parse the way html5lib/filters/sanitizer.py does — from six.moves import urllib_parse as urlparse, a RENAMED re-export through two shims — because testing urllib.parse directly would pass without exercising that path, and that path is the one that did not work before v351. 27 checks, identical under CPython.

The cases are the sanitizer's own, not generic ones: the scheme is lower-cased (a lower-case allow-list otherwise lets JavaScript: through — a security difference, not a cosmetic one), data: keeps its semicolon while http: splits ;params off (the sanitizer reads uri.path to check a data: content type), and an unbalanced [ raises ValueError (the sanitizer catches exactly that and DELETES the attribute).

Score: past the wall 0, onto the next wall 5

missing module: six_moves 5 → 0.

compile 6/48 → 6/48. yield is now 18 files, and this ticket put four of them there.

Attribution: the row cleared because this shim exists — before it, those files stopped at missing module: six_moves. But sanitizer.py also needed frank2's rename fix to get past urllib_parse as urlparse, so that file is a joint result and not this ticket's alone.

The one file this ticket existed for moved one step, onto a new wall

sanitizer.py:769 is super(Filter, self).__init__(source) — the two-argument super, which does not parse: [[bug-n-two-argument-super-does-not-parse]] (N, p60). html5lib writes it in every filter, so it is not one file's habit.

Still out of scope, unchanged

The http_client half. _inputstream.py imports it and would stop there if yield were fixed; lib/rtl/http.pas exists and needs a Python face — [[feature-b-mimic-urllib-request-over-the-rtl-http-stack]].

One workaround taken

ParseResult.__getitem__ binds its field tuple to a local before indexing, because subscripting a container literal inside a function body segfaults ([[bug-n-subscripting-a-container-literal-inside-a-function-segfaults]], N, p65, found here). Registered in devdocs/dev/track-b-workarounds.md with its revert condition.

Log

• 2026-08-18 — resolved, commit 53ccaa4da.