[d-torrance]

Three checks on user input were in the wrong place: two aborted a Debug build
on input a Release build reported and declined, and the third sat where no
test could reach it. PR #66 and PR #69 settled the reasoning for the first
shape; PR #83 said the third would want its own PR.

-divisorLookup and -monomialTable

staticMonoLookupMake's switch asserted before it threw, and nothing
validated the code earlier:

$ mgb gb cyclic5 -divisorLookup 99      # Release
Unknown code for monomial data structure

$ mgb gb cyclic5 -divisorLookup 99      # Debug
mgb: StaticMonoMap.hpp:431: Assertion `false' failed.
Aborted (core dumped)

Deleting the assert would make the two agree, but it would leave the value
unchecked until it reached a switch four call levels down. PR #69 faced the
same choice and took the other route — it put the modulus check in
F4Reducer's constructor and left the asserts downstream, noting that
"establishing the bound here lets [them] assert it". So this checks the code
at the two entry points that take it from outside, MonoLookup::makeFactory
and ModuleMonoSet::make, and leaves the switch alone.

The error arrives when the factory is made rather than when it is first used,
and it lists the codes:

ERROR: Unknown code 99 for the monomial data structure.
The supported codes are:
  1   list, using divmasks
  2   KD-tree, using divmasks
  3   list
  4   KD-tree

-monomialTable reaches ModuleMonoSet::make, a second path with the same
abort, and is covered too.

The streaming interface

MATHICGB_STREAM_CHECK asserted before it threw at twenty sites in the public
streaming interface, so a Debug build aborted on exactly the caller protocol
violations the macro exists to report — calling idealBegin() twice, and so
on. Nothing had ever caught it: every test that drives an ideal stream follows
the protocol, so no check had fired, and before PR #80 no CI cell built this
code with assertions at all.

This also removes the 19 clang warnings PR #81's CI turned up in the macOS
debug cells. They came from the assert(("message", condition)) idiom, whose
left operand genuinely has no effect; deleting the assert removes the idiom.

The matrix file's modulus

PR #74's primality check lives in src/cli/MatrixAction.cpp, which is the one
place it cannot be tested — mathicgb-unit-tests links no src/cli object
and there is no CLI-level harness.

It moves into SparseMatrix::read, beside the size check PR #83 added, where
the modulus is still intact and a test can reach it. QuadMatrix::read needs
nothing of its own: it reads its four submatrices through
SparseMatrix::read, so all four are checked. The move also covers a read the
CLI never checked at all — the reference .rbrmat that a second run compares
its output against.

What stays in the CLI is the file name, which read cannot know, having only
a FILE*:

$ mgb matrix composite.brmat
While reading composite.brmat:
ERROR: The modulus 100 is not prime. MathicGB only supports prime fields.

PR #83's oversized-modulus error gets the file name the same way.

Four commits

1. 41dce00 — MonoLookup::displayCodes and ModuleMonoSet::displayCodes
   had byte-identical bodies; both now read one table next to the switch that
   enumerates the same codes. No functional change.
2. ac4054b — validate the code at the two entry points, with
   src/test/MonoLookup.cpp covering both.
3. ecd68c4 — MATHICGB_STREAM_CHECK throws without asserting, plus the
   first test to exercise any of those twenty checks.
4. 2f12500 — the primality check moves into SparseMatrix::read, with a
   test at each entry point.

Each builds and passes standalone.

Verification

• 253/253 in cmake Release and Debug, no build warnings from GCC 13.3.
  Test count 248 → 253.
• autotools make distcheck passes.
• clang 18.1.3 on src/mathicgb.cpp: 19 warnings before, 0 after.
• Each new test fails without the check it covers.
• mgb gb -divisorLookup 99 and mgb sig -monomialTable 99 now report
  identically in Debug and Release.
• mgb matrix refuses a composite-modulus .brmat with exit 255; a prime one
  reduces and writes its output as before.

One thing this does not fix

mgb gb -monomialTable 99 still exits 0. That option is offered by
GBCommonParams, which both actions share, but only SigGBAction reads it,
so for the gb action any value is accepted because none is used. That is a
separate defect from the one here and wants its own change.

🤖 Generated with Claude Code