Skip to content

preprocessing.md

Latest commit

 

History

History
223 lines (178 loc) · 9.37 KB

preprocessing.md

File metadata and controls

223 lines (178 loc) · 9.37 KB
Raw
title Preprocessing Stage
audience developers, maintainers, contributors
prerequisites contributor architecture guide, native project compiler flags
related ../architecture.md, index.md, parsers.md, semantics.md
status maintained
publication reviewed

Preprocessing Stage

Purpose And Boundaries

prik/preprocessing/ turns original Fortran source into parser input and measures the compiler-dependent facts that semantic conversion needs. It owns compiler invocation, source provenance, native INCLUDE expansion, and target probes. It does not parse declarations, construct semantic IR, choose semantic scalar identities, or complete wrapper policy.

The package contains early C-frontend modules: c.py collects raw directive metadata and probes/c_types.py measures C ABI facts. C support is not yet complete; a future C frontend may build on them. They do not participate in the current Fortran wrapper path.

A Fortran Source Through This Stage

The primary route is preprocess_source() in source.py. It retains the prepared source and the facts that explain how it was produced:

Fortran source path + PreprocessingConfig
  -> select compiler command: direct compiler, compile database, or template
  -> compiler-expanded source
  -> source mappings, dependencies, optional macros, diagnostics, and recipe
  -> native Fortran INCLUDE expansion
  -> PreprocessResult.source
  -> Fortran parser

semantic requirements + compiler and target configuration
  -> Fortran type probe: generate, compile, and run a small program
  -> FortranTypeProbeReport(values, recipe, source_text)
  -> Fortran-to-IR conversion

PreprocessResult is prepared parser input plus provenance; it is not parsed syntax or semantic meaning. FortranTypeProbeReport is a compiler measurement plus its recipe; it is not a stable semantic scalar or NumPy dtype.

Local Structure

prik/preprocessing/
├── __init__.py
├── source.py
├── fortran.py
├── c.py                         deferred C inspection support
└── probes/
    ├── fortran_types.py
    └── c_types.py               deferred C inspection support

Directory Tour

Module Public boundary and result Change it when
prik/preprocessing/__init__.py Re-exports the supported shared source-preparation records, adapters, and entrypoints. The shared preprocessing import surface changes.
prik/preprocessing/source.py preprocess_source() is the compiler-backed route. PreprocessingConfig selects its command; PreprocessResult returns expanded text, provenance, and diagnostics. Compiler adapters, invocations, recipes, mappings, dependencies, macros, or diagnostics change.
prik/preprocessing/fortran.py expand_native_fortran_includes() turns remaining textual INCLUDE statements into parser input while retaining mappings and diagnostics. Native Fortran include discovery or expansion changes.
prik/preprocessing/probes/fortran_types.py evaluate_fortran_type_requirements() and evaluate_fortran_type_facts() turn semantic requirements into cached compiler measurements; FortranTypeProbeReport retains values and recipe. Fortran fact generation, validation, cache identity, or semantic-facing probe results change.

Module Workflows

source.py: compiler-backed parser input

preprocess_source() is the normal boundary. Start with it when changing the prepared-source path. It validates PreprocessingConfig, then build_preprocess_invocation() selects a command-template, compile-database, or direct-compiler invocation. It executes that invocation, collects compiler line markers, dependencies, optional macros, and diagnostics, then delegates remaining native Fortran INCLUDE statements to fortran.py.

The public records divide the work: PreprocessingPlan and Invocation describe a request and command before execution; PreprocessingRecipe and PreprocessResult record the completed operation; SourceMapping, IncludedFile, MacroDefinition, and PreprocessingDiagnostic retain its side facts. The convenience runners return only source or source plus a typed recipe; use preprocess_source() when the next stage needs the complete result.

The source is organized in the same order: configuration and validation, adapter facades, invocation construction, provenance recovery, then execution and result assembly. Private helpers belong to one of those phases.

fortran.py: native includes after compiler expansion

expand_native_fortran_includes() receives an already compiler-expanded Fortran stream. It resolves an INCLUDE beside its including source before the configured directories, expands it recursively, and returns parser text, dependency edges, source mappings, and diagnostics. It records a missing file or cycle rather than making parser decisions; source.py promotes error diagnostics after assembling the complete result.

probes/fortran_types.py: target facts for semantic conversion

The module generates a small Fortran program, compiles and runs it for the configured target, validates its result, and caches the report by compiler, flags, runner, environment, and generated source.

evaluate_fortran_type_requirements() and evaluate_fortran_type_facts() receive semantic requirement records. They reuse a supplied or cached FortranTypeProbeReport and return the values that the Fortran semantic converter needs. probe_fortran_type_expressions() is the uncached measurement boundary; probe_fortran_type_expressions_cached() is the normal repeated-use boundary. A report is valid only for the target identity encoded in its recipe and cache key.

Run The Workflows

source.py demonstrates the source-preparation handoff. Its complete direct example also prints a small C source-preparation demonstration; C frontend support remains future work.

python3 prik/preprocessing/source.py
Before Fortran include expansion:
module greeting
include 'constants.inc'
contains
subroutine show_answer()
print *, answer
end subroutine show_answer
end module greeting

After Fortran include expansion:
module greeting
integer, parameter :: answer = 42
contains
subroutine show_answer()
print *, answer
end subroutine show_answer
end module greeting
Native includes: 1; diagnostics: 0
...

The script supplies a small module with one include and prints the parser input before and after preprocessing. The replacement of the include line, plus the dependency and diagnostic counts, shows that source expansion preserves the information needed by the parser without reporting a problem.

The Fortran-specific module shows the parser input and provenance it returns:

python3 prik/preprocessing/fortran.py
Expanded parser input:
module geometry
integer, parameter :: dimensions = 3
end module geometry
Native include dependencies: 1
Generated source mappings: 5
Diagnostics: 0

This narrower example prepares one Fortran source with one native include. Its expanded text is the parser input; the mapping count records how generated lines still lead back to their source locations.

The target-probe example shows a compiler measurement, not a fixed semantic type. It requires gfortran or f95.

python3 prik/preprocessing/probes/fortran_types.py
selected_int_kind(9) = 4

The reported number is target-dependent. The example establishes that the compiler, rather than PRIK, supplied the fact.

Tests And Evidence

Evidence What it establishes
Fortran preprocessing Adapters, recipes, mappings, native includes, diagnostics, and parser handoffs.
Parser boundaries Prepared source reaches parsing with preserved facts and unsupported raw constructs stop at the correct boundary.
Fortran type probes Compiler facts, requirement evaluation, cache separation, and report validation.

Change Routes

  • Change compiler expansion, commands, provenance, recipes, or diagnostics in source.py.
  • Change native Fortran INCLUDE behavior in fortran.py.
  • Change compiler-measured Fortran facts or cache identity in probes/fortran_types.py.
  • Change stable scalar identity in semantics/, backend dtype projection in codegen/, and wrapper behavior in the later owning stage.

Boundaries And Invariants

  • Preserve original coordinates through every source transformation.
  • Keep compiler expansion, native include expansion, and target measurement as separate operations.
  • Run probes in temporary directories so compiler products cannot enter the repository.
  • Do not reuse source provenance or a probe report across a materially different compiler target.

Failure Boundary

This stage reports invalid preprocessing configuration, compiler execution failures, missing native includes, include cycles, and invalid probe results. It delegates declaration syntax to parsers/, stable meaning to semantics/, and wrapper support to later stages. Start with the first incorrect prepared source, provenance record, or compiler fact—not the later parser or build failure.