Writing FRD#

pyvista_frd.write writes a PyVista mesh as long ASCII, binary float32, or binary float64 FRD. pyvista_frd.convert changes an existing document’s encoding without building a mesh. With no target encoding, conversion preserves all four FRD encodings, including short ASCII.

import pyvista_frd

mesh = pyvista_frd.read("result.frd")
pyvista_frd.write("copy.frd", mesh)
pyvista_frd.write("result_binary.frd", mesh, binary=True)

pyvista_frd.convert("result_binary.frd", "result_ascii.frd", binary=False)

Files created by this package identify pyvista-frd-reader as the writer. They are not labeled as CalculiX output.

Byte round trips#

A writer and reader from the same implementation can share a defect and still pass a write-then-read test. This project therefore checks original bytes from external files in addition to semantic round trips.

tools/sweep_rewrite.py reads each document, writes it in its original encoding, and compares every byte:

Corpus	Files with FRD blocks	Byte-identical
CalculiX regression suite	834	834
GitHub code search, `extension:frd`	277	277
Total	1,111	1,111

Another 655 files contain no FRD node, element, or result blocks. The converter copies these files unchanged and reports them separately. The sweep fails if it finds no file with FRD blocks, so an empty or irrelevant corpus cannot pass.

tests/test_writer.py applies the same byte check to the 24 CalculiX-generated fixtures. Handwritten fixtures use non-CalculiX number formatting, so their tests compare the resulting mesh and values by bit pattern instead.

Cases found by the external corpus#

CalculiX GraphiX record layout#

Two cgx_2.x/examples/result.frd files use a different ASCII layout:

 -1    1 0.86430E-07-0.16898E+02 0.55880E+02      cgx example
 -1         1 8.64300E-08-1.68980E-01 5.58800E+01  common ccx layout

The GraphiX files use five-column ID fields, 15 element IDs per line, and Fortran E12.5 formatting. Their headers do not include a format code. The writer infers record width and number formatting from the first record of each block and preserves that layout.

Binary-to-ASCII rounding#

CalculiX casts each value to float32 before formatting ASCII. This writer formats the stored float64 value. The outputs differ only at rounding ties.

Across 12 paired binary and ASCII fixtures, 800 of 825 record lines match. The other 25 lines differ by one final digit. Those 25 lines contain 0.8% of the 3,044 values.

This choice preserves the value held by a float64 binary block or NumPy array. It means a binary-to-ASCII conversion is not always byte-identical to ASCII written by CalculiX. Same-encoding round trips are unaffected. See Differences from PyVista for the exact case.

Adjacent element IDs#

Five-column fields can contain adjacent five-digit IDs:

 -21000110002100031000410005100061000710008

Whitespace splitting treats this as one integer and loses the connectivity. The writer parses these records by fixed width. The case is pinned by test_glued_face_ids_survive_the_writer.

Cross-encoding check with CalculiX#

Same-encoding byte comparison cannot detect errors that occur only during conversion. Binary blocks, for example, have no -3 terminator; ASCII blocks require one.

The CalculiX regression suite includes a submodel that reads beampdouble.frd.ref. The following inputs were tested:

Input	CalculiX result
Original binary file	Accepted; baseline output
Binary file rewritten by this package	Accepted; output matches baseline byte for byte
Binary converted to ASCII without a terminator	Rejected
Binary converted to ASCII after the terminator fix	Accepted; output matches CalculiX’s ASCII reference

The final comparison excludes four header lines containing dates and CalculiX versions. Ten coordinate values also differ in their final digit unless this writer’s values are first narrowed to float32, as described above.

Coverage limits#

A newly constructed mesh has no original FRD document for byte comparison. Its records use the same emitter, while its headers are covered by C++ writer tests and write-then-read tests.
Format 0 (short ASCII) and format 2 (binary float32) appear in none of the 1,766 external files. They are covered by project write-and-read tests.
External binary coverage contains one published format-3 file. Another 12 format-3 files are generated from project decks with CalculiX.
Two external files contain NaN. The writer uses a fixed spelling instead of platform printf output.
Byte identity proves preservation of the document, not the meaning of fields ignored by both CalculiX and this implementation.

The observed format-code counts are:

Format code	Blocks	Files
0, short ASCII	0	0
1, long ASCII	2,138	1,070
2, binary `float32`	0	0
3, binary `float64`	1	1
No code	4	2

Mutation checks#

tools/mutate.py applies 10 writer defects and verifies that the named test fails for each one:

Mutant	Expected check
`writer-value-precision`	Byte round trip
`writer-narrows-through-float32`	Binary-to-ASCII comparison with CalculiX
`writer-faces-wrap-at-nine`	Byte round trip
`writer-values-wrap-at-five`	Byte round trip
`writer-face-width-assumed`	Byte round trip
`writer-faces-split-on-whitespace`	Adjacent-ID test
`writer-converts-without-a-terminator`	Terminator test
`writer-truncates-array-names`	Write-then-read test
`writer-reuses-the-readers-permutation`	C++ `WriteTest`
`writer-hoists-unparsed-lines`	Stray-line position test

All 10 are rejected by their expected checks. The stray-line test was added after that mutant initially survived, and it covers both node and result blocks.