33 Commits

Author	SHA1	Message	Date
lilleman	3ae8b9f1b1	Test that New without WithSeed fails naming it, that the CLI without --seed draws afresh each run, and seed every test's New	2026-10-12 18:34:15 +02:00
lilleman	3368f085fd	Test that the shipped data loads from a Go package per module, that a bare New fails naming WithDataFS, and that a read nothing provides names the modules its reader's manifest reads	2026-10-11 16:02:48 +02:00
lilleman	55d503317d	Answer architecture round 3: New loads only sources without an index and what they read, an entry drops reads, ErrLoad sits beside New, the manifest decoder moves to manifest.go and names each mistake once as jsonvalue.Kind words it	2026-10-10 11:24:32 +02:00
lilleman	3c1af91433	Test that a manifest refuses every key and shape it does not define, that a table under a shipped table loads from a data path, that an omitted parent fails the first reach, and pin New's allocations	2026-10-10 11:24:32 +02:00
lilleman	63223c6d63	Answer stability round 1, prose round 1 and product-owner round 1: startOf's doc names the head, the old senses leave names, comments, messages and tests, and the vocabulary defines scope, link and head	2026-10-07 01:04:44 +02:00
lilleman	fb3fb16deb	Answer stability round 3: the allocation case of a read two levels under a pick holds at its baseline with no margin, as the performance gate decision now says	2026-10-06 13:46:44 +02:00
lilleman	f07764bce1	Hold the allocation case of a read two levels under a pick at 12, what the lowest supported Go allocates	2026-10-06 13:46:44 +02:00
lilleman	c814d36985	Answer prose round 1: pickKey names its two roots, readUnder says when it reads, levelKeys, draw and readField read on one pass, and the allocation case is named for what it reads	2026-10-06 13:46:44 +02:00
lilleman	93c96968dc	Answer stability round 1: readUnder builds the leaf key once, the pick's comments name the pickKey, and the allocation gate covers a read under a nested pick	2026-10-06 13:46:44 +02:00
lilleman	1a15a139c8	Answer architecture round 1: refuse a name read inside the field it binds, check every scope's cycles, build a column read in one pass, drop the implicit struct pick for names across tags, and state the fresh-draw rule once	2026-10-05 10:19:01 +02:00
lilleman	5a3abc325f	Hold the allocation gate to what both supported Go versions allocate	2026-10-05 10:19:01 +02:00
lilleman	5b23ca7706	Test that every {…} draws afresh and only a name keeps a pick, a field's and a struct's path tags included	2026-10-05 10:19:01 +02:00
lilleman	05cf531339	Test a name read across draw groups or in a repeat, read once through a path, and misspelt	2026-10-03 14:23:58 +02:00
lilleman	30d18bf485	Test one name across draw groups, two names on one category, and a repeat of a name's allocations	2026-10-03 14:23:58 +02:00
lilleman	b99966fe19	Name the render's draws apart from the expansion hold: drawMemo, groupDraws, renderDraws	2026-09-29 14:45:34 +02:00
lilleman	0225f16924	Name the tests' holds, header and row lookup by their one meaning, and drop the dead bound_test.go pointer	2026-09-28 12:06:03 +02:00
lilleman	8c856b6510	Want a column of a drawn row read without the probe allocating	2026-09-28 00:55:48 +02:00
lilleman	203b235fdd	State the copy rule on renderScope once, and say the allocation gate prices what keeps a hold set off the heap	2026-09-26 15:17:05 +02:00
lilleman	cbe982f52b	Price two held paths at what the lowest supported Go allocates	2026-09-24 13:24:58 +02:00
lilleman	c77a2e2c93	Gate the allocations of paths read from one held draw	2026-09-24 13:24:58 +02:00
lilleman	41dbbd3fda	Say hold where the prose still said draws	2026-09-21 22:06:01 +02:00
lilleman	fb83aa6178	Test the family fence over replayed selectors, a name without a key, a name shadowing a key, a table reading its own family, an ancestor's column clash, a BOM header and five-deep pins	2026-09-17 12:45:17 +02:00
lilleman	d4060e041a	Test the table node: rows TSV, key and name selection, linked tables drawn consistently, and the choice-of-rows fence	2026-09-17 11:47:39 +02:00
lilleman	b45a1edae0	Cut the comments that restate the README or narrate the arrangement, and name the record fence asymmetry in Decisions	2026-09-16 10:43:25 +02:00
lilleman	0443ecdf13	Tests: gate the allocations a reference path costs across a repeat and in a named draw group	2026-09-16 10:29:00 +02:00
lilleman	d4ff3b242e	Tests: FakeStruct fills tagged fields as one record, draws a nested struct apart, refuses a tag its field cannot hold, and compiles a type once	2026-09-15 16:23:00 +02:00
lilleman	1ad9418a76	Tests: a printed zero is no divisor and prints unsigned, int64 refusals say so, and CSVLine allocates twice	2026-09-15 15:08:14 +02:00
lilleman	5aec9f19fc	Tests: name the record entry points after the string ones	2026-09-14 23:37:25 +02:00
lilleman	2310cb1e1a	Name every spelling the self-reference fence refuses, and cut the doubled comment	2026-09-04 09:53:46 +02:00
lilleman	b99526aeb8	Tests: a record read whole or as an operand still points back at itself, and Record under concurrent use	2026-09-04 09:46:39 +02:00
lilleman	79827a66d6	Fence a record once per node, refuse a column reading its own record, and keep a bare reference independent as an operand	2026-09-04 09:32:29 +02:00
lilleman	ce22c83e7d	Cut the review-found narration: two helper invariants, one line each	2026-09-03 21:53:51 +02:00
lilleman	296058fb80	Tests: allocation ceilings for deeply nested and wide renders	2026-09-03 21:46:22 +02:00