graph: refuse unresolvable and non-directory parents

[typeless]

Closes #492.

Four commits: #492 itself, plus the similar problems its review found.

1. add_file_node refuses an unresolvable parent id (#492)

A non-zero parent_dir that get_file_node cannot resolve is now InvalidNodeId, the code add_edge already returns. Before this, such a node was filed under the source root, and its name went into another node's dir_children bucket or past the end of dir_children. The new test observed that out-of-bounds write as a SIGSEGV. No production caller reaches this error.

2. glob: match files, never directories

glob_expand added any directory whose name matched, and the builder then bound it as a File node. Upstream's nl_add_path globs through tup_db_select_node_dir_glob with directories excluded. Red: gdir_* → { "gdir_sub", "gdir_top.txt" }. Now it returns only the file. New requirement: REQ-GLOB-FILES-ONLY.

3. graph: refuse a path under a node that is not a directory

Before this change:

• : sub |> … followed by an output sub/x.o ran the command, which failed in the shell.
• Two rules writing gen and gen/x both ran, and one failed.
• A group sub/<g> under a file built with no diagnostic at all.

Upstream refuses all three in find_dir_tupid_dt_pg. add_file_node now refuses any parent that is not a Directory or GeneratedDir, with a message like Unable to create 'sub/x.o' because the file 'sub' is not a directory. New requirement: REQ-OUTPUT-UNDER-DIRECTORY.

The check exposed two producers that typed directories as files:

• the glob (commit 2): putup's own test runner globs | ../e2e/fixtures/** and was refused;
• resolve_input_node: a real directory named as a rule input became a File node, and a reviewer found six such Tupfiles that tup accepts and the check refused. It now asks the disk.

4. Delete resolve_phi_node

Nothing calls it, and its INVALID_NODE_ID fallback is also SOURCE_ROOT_ID.

Evidence

• Each fix has a red-then-green test, and each red was observed before its fix.
• make check → all tests passed (182569 assertions in 959 test cases, 32 e2e shards).
• Bootstrap scripts regenerate unchanged.
• Fable reviewed on two lenses, regression and correctness. Its 20 Tupfile fixtures exit identically under the installed pre-change binary and this branch.
• The helloworld, busybox and bsp examples parse unchanged.

Not fixed here

• Order dependence (Refusing an output under a non-directory depends on rule order, because putup learns a path's type only when a rule names it #495). putup learns a source path's type only when a rule names it. An output named under a file before any rule reads the file creates a Directory node and still fails in the shell. Relatedly, ensure_file_node(Generated) returns an existing Directory unchanged, where tup reports "Attempting to insert … as a generated node".
• Accessor defaults (Graph accessors answer an unresolved id with a value a real node can have, because INVALID_NODE_ID is the source root #496). get_parent_dir returns 0 for an unknown id. That is the family's shared contract, not one function's defect: get<NodeType> returns File, get<NodeFlags> returns None, and so on. The root cause is that INVALID_NODE_ID == SOURCE_ROOT_ID. All production callers pass live ids.
• is_guard_satisfied left as is. All 14 callers pass live command ids, and "not a command, so no guards" is a correct answer.

🤖 Generated with Claude Code

https://claude.ai/code/session_01StgwMENEyfBnEoe4pvAdtQ

[github-actions]

PR metrics

Performance (gcc example, Linux)

Workload	Instructions	CPU time	Page faults	D1 miss	LL miss	Wall	Peak RSS
parse	1760 M (+0.1%)	0.64 s	15.2 k	0.7% (+0.1pp)	0.1% (+0.1pp)	0.655 s	35.1 MB (-0.1MB)
dry-run	2486 M	0.75 s	16.8 k	0.8% (+0.2pp)	0%	0.772 s	41.3 MB (-0.1MB)

Deterministic signals: instructions (cachegrind-simulated instruction reads — exact across runs, no PMU needed), page faults, peak RSS, and the cachegrind D1/LL miss rates. CPU time is user+sys from time(1).

Internal statistics (gcc example, up-to-date dry run)

Metric	Value
Tupfiles parsed	24
Commands	3545
Commands scheduled	0
Files checked	5834
Files changed	0
Files in index	6188
Graph edges	384927
Index size (bytes)	7869819
Implicit deps	344126
Hash computations	197
Hashes skipped (stat cache)	5636
Stat calls	5885
Parse time (ms)	579.1
Total time (ms)	756.2
Runner CPU	AMD EPYC 9V74 80-Core Processor

Counters from putup -n --stat on the fully-built gcc example (up-to-date dry run): deterministic work measures — a jump in commands scheduled, hash computations, or stat calls is a real behavior change, not noise. Timings are the minimum over repeated runs, compared only against a baseline from the same CPU model; the counters are the regression signal.
Timing deltas suppressed: baseline ran on different hardware (INTEL(R) XEON(R) PLATINUM 8573C).

Binary size (Linux)

Binary	.text	.data	.bss	File
putup	603.2 KB (+0.1%)	2.4 KB	98.8 KB	716.3 KB (+0.6%)

Code churn (whole codebase, last 30d)

Files	Lines written	Still present	Churned	Churn rate
32	2522	2185	337	13.4%

Of the lines written across the codebase in the last 30 days, how many are already gone — work that was written and then discarded or rewritten inside the same window. This is the state of the tree including this PR, not a measure of the PR itself. Only code we write is counted: tests, examples, vendored and generated files, CI plumbing and prose are excluded. 1639 lines were deleted in the window in total, most of them older than it.

Where the churn is

File	Lines written then discarded
src/parser/eval.cpp	77
src/graph/builder.cpp	72
src/graph/dag.cpp	56
src/index/entry.cpp	44
include/pup/core/token_list.hpp	23
include/pup/parser/eval.hpp	16
src/cli/cmd_build.cpp	8
src/index/reader.cpp	8
include/pup/core/instruction.hpp	7
src/core/instruction.cpp	7

Test coverage (lines)

Overall	Median file	Min file	Max file
88.9% (+0.1pp)	96.5% (-0.2pp)	14.7% include/pup/parser/token.hpp	100.0% include/pup/core/arena.hpp

105 files · 17915/20154 lines covered

Deltas vs main@b595d6c70.

Updated for 1fd6b5a