65 Commits

Author	SHA1	Message	Date
lilleman	2a7499e93c	Parse and check a format in one call, keeping compileTemplate under the complexity gate	2026-09-24 23:36:27 +02:00
lilleman	467cb6fb18	Parse a format once into tokens, and have the fences read its ops	2026-09-24 23:36:27 +02:00
lilleman	2e778f3b31	Name a cell's line in its link-time errors, and say where a format holding a reference compiles	2026-09-24 22:45:36 +02:00
lilleman	526832fc3f	Compile a format holding a reference once, at link	2026-09-24 22:45:36 +02:00
lilleman	5aae86638b	Keep reference bindings out of template.fields, in template.refHeads	2026-09-24 19:38:10 +02:00
lilleman	9f00ad039e	Move the decisions to docs/decisions.md, indexed in AGENTS.md and linked from the fences they back	2026-09-24 13:44:06 +02:00
lilleman	def2a5f25d	Gather the family fence in family.go	2026-09-23 23:11:13 +02:00
lilleman	41dbbd3fda	Say hold where the prose still said draws	2026-09-21 22:06:01 +02:00
lilleman	d128621fc1	Name a render's holds apart from the draw fence	2026-09-21 22:06:01 +02:00
lilleman	a2074a455e	Put the reference write beside fields, and cut the header restatement	2026-09-21 11:43:58 +02:00
lilleman	311ed40d05	Say which pass fills the cell fields, and when the format recompiles	2026-09-21 11:43:58 +02:00
lilleman	1e80fd54f6	Group the template fields under the pass that fills each	2026-09-21 11:43:58 +02:00
lilleman	ce8ea550a3	Correct what the readers proved false, and lift the file map out of last place	2026-09-20 21:47:17 +02:00
lilleman	fb0480e741	Identify a fence alternative by the table row rather than the cell, keep the outermost row through a reached table, and key reads and visits by it	2026-09-17 15:53:44 +02:00
lilleman	7abd3e2d95	Refuse a blank TSV line, fence a table's family through any template, resolve a path before drawing, clamp the weighted pick, and say which countries and currencies the tables hold	2026-09-17 13:15:33 +02:00
lilleman	f6b54c8ae6	Add the table node: rows TSV, key and name selection, linked tables drawn consistently, and the choice-of-rows fence	2026-09-17 12:28:46 +02:00
lilleman	68d2461704	Name the draw option drawGroup, refuse one whose reads all sit behind a nested draw group, and record why the two fences stay apart	2026-09-16 08:55:55 +02:00
lilleman	d3f430bf53	Render a repeat iteration in no group, refuse a group that names nothing, key group names by category, and own reference overlap in one fence	2026-09-15 23:31:48 +02:00
lilleman	5ed45cc55c	Rename the folder node type from group to folder, freeing group for the draw option	2026-09-15 23:25:39 +02:00
lilleman	246f99a319	One reference draw per render in both views: a repeat iteration draws anew, a group option draws apart, and one fence refuses what a render's draws cannot answer for	2026-09-15 22:59:30 +02:00
lilleman	b4ebab5abf	Keep a read's null with its draw, share the lone-reference check, name a JSON-string item's object form again, prove each column once per pass, and read a record off template.record	2026-09-15 20:42:13 +02:00
lilleman	2b2010bdd6	Take the datatype and null of the column a lone reference reads, and check a column's datatypes once its references are linked	2026-09-15 20:20:25 +02:00
lilleman	3cd3f8f317	Spend the struct cap on embedded structs too, word it for any struct field, and route path advice through the name check	2026-09-15 17:50:42 +02:00
lilleman	bbe6c5b1d9	Fold the range check into columnKind.past, refuse a leading / in IsTemplate, reserve the name -, and document fake:"-" on an embedded struct	2026-09-15 17:11:20 +02:00
lilleman	3e7127da71	Parse a divisor's rounding half exactly, print zero unsigned, name int64 refusals, and quote CSV fields without a writer	2026-09-15 15:08:26 +02:00
lilleman	5615ab6d89	Check a typed column's value, not its text: one literal, value call, calc or read, bounded by its arguments	2026-09-15 14:33:04 +02:00
lilleman	94dc562c6c	Typed columns and null: a datatype option, null items, and a load check that every typed render parses	2026-09-15 11:23:57 +02:00
lilleman	1cd422352b	Cut the comments the README already records, and record the cycle-fence and exit-code rules there	2026-09-03 21:00:42 +02:00
lilleman	44c91668f9	Give the inline pipeline its own file, and record the shape, padding and template-API decisions	2026-09-03 20:37:44 +02:00
lilleman	6916894ab5	One CLI classifier settling the argument shape before load; JSON strings are templates and a quote is reserved in names	2026-09-03 20:34:42 +02:00
lilleman	873b7678ba	Address reviewer findings: reference hint, consumer-terms type errors, misuse exit code, immutability and parity docs	2026-09-03 19:21:42 +02:00
lilleman	305d8f8edd	Address stability nits: doc grammar, scalar-JSON note, bracket-rule rationale	2026-09-03 17:53:05 +02:00
lilleman	d8ece597c8	Reserve brackets in names, add compile-once NewTemplate, let --repeat reuse it, record scope artifacts	2026-09-03 17:48:20 +02:00
lilleman	3bc3e35045	Add FakeTemplate: compile an inline template and render it against the loaded tree	2026-09-03 17:27:54 +02:00
lilleman	addba9abf7	Give maxLen its own literal and repeats MaxRepeat; arg errors name the spelling and the range; write's comment says what Flush reports; assert 64-bit at compile time	2026-09-02 19:25:19 +02:00
lilleman	c5304bb6e7	Record the repeat cap's reach, 64-bit only, the constant-divisor rule and the rejected default spellings; correct the repeat cap comment	2026-09-02 19:09:50 +02:00
lilleman	4f9285ab5c	Reject non-finite float bounds, unplain integer args, constant samples, a constant zero divisor and the default separator; a binding key is no path segment; ascii keeps DEL; an unheld path panics	2026-09-02 19:08:32 +02:00
lilleman	f9647ff6bb	Decompose compileTemplate, checkBoundLevelsHeld and loadDir; gate cyclomatic complexity at 14	2026-09-02 18:31:38 +02:00
lilleman	bcbf044290	One path walk under checkPath, descend, readField, pathLeaves and coverPath; split the hold and the render graph into their own files	2026-09-02 18:29:45 +02:00
lilleman	1bf5470c6f	Reference sigils follow the filesystem: / the root, . this folder, .. the folder above	2026-09-02 18:26:02 +02:00
lilleman	1b1b93b35f	Reject weight 0, a never-numeric calc operand and a repeat product past the cap; the CLI classifies ErrNoData	2026-09-02 18:23:03 +02:00
lilleman	c5c06375d8	Reject a repeated bare token of a held name, naming the single-token spelling	2026-09-02 18:21:03 +02:00
lilleman	8cd603a2b1	Key the repeated-item check on the string itself; marshal only objects	2026-09-02 12:59:12 +02:00
lilleman	31c2f12098	Hold a reference path like a sibling path; add the lowercase, uppercase and ascii transforms	2026-09-02 12:52:19 +02:00
lilleman	054f99c94f	One spelling per shape: reject a one-item choice, a repeated item and an inert object; reshape the data	2026-09-02 12:38:59 +02:00
lilleman	ff4b243148	Format text is literal; class runs are {digits(n)} {upper(n)} {lower(n)}; a string is a format	2026-09-02 12:37:21 +02:00
lilleman	05dc042e49	Rename the project to fejkdata and record the fork source	2026-09-01 20:24:03 +02:00
M	64a0d29496	Say why an empty name is rejected, and finish what that removed ... The reason given was wrong. A dot path did reach an empty-named field: Fake("a.") returned it, Fake("a.b.") returned a nested one, and {..a.} bound it. What was true is narrower — List never offered it, because an empty name is no path segment — so the engine accepted spellings it would not advertise. The message, the test comment and the README say that instead. Two things the rejection finished off: A {} token still reported "no field \"\"", pointing at a fix the loader now rejects — two errors for one rule. It is told the name can never exist, like an option token already is. addressable is dead: both halves of "not empty and no dot" are now rejected where a name is authored. Mutating it to panic leaves the suite green here and panics on main, so it was live and is not. Keeping it would preserve the silent-hiding this change removes — List quietly omitting a name rather than New refusing it. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-01 07:31:18 +02:00
M	138bd21d83	Reject an empty field name ... A field named "" was reachable by exactly one spelling, a bare {} token, and by no dot path: List hid it and Fake could not ask for it. checkName rejects it now, for the reason it rejects a dot — a name no path can reach. That also makes the empty-segment rule unconditional. {a.} was rejected because a field really named "" would otherwise have made it resolve; no data can carry one, so the rule now guards a shape that cannot exist. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-09-01 07:31:18 +02:00
M	909596ec2d	Trim comments that restate or argue ... The operand order was written out in three places; it now lives at calcVars, which fixes it, and the other two point there. Dropped the note explaining why ')' is not reserved, which said what the set is not rather than what it is, and shortened reservedList's doc to less than its body. The note on held now states where it reaches instead of arguing against a design this repo no longer contains. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>	2026-08-31 21:58:37 +02:00