tclpdf

What a licence audit of tclpdf found
Login

What a licence audit of tclpdf found

1. Why and what

tclpdf was written with Claude Code (see the provenance note, cb5371ca27). A package made with AI assistance has to show that it carries neither code taken from other projects nor code under a copyleft licence, so the author had release 1.7 audited for origin and licences; this note of 2026-11-01 reports the result. The object was the checkout of tag version-1.7 (check-in 9ba4e3def9, 2026-10-16), 516 versioned files (fossil ls -r version-1.7). tclpdf is pure Tcl: the repository holds no .c, .h, .cpp, .js or .py file and no vendored library; the one non-Tcl source is tools/formcheck.java, a test helper of make check that uses PDFBox at run time and is not shipped.

2. How it was checked

Check Scope Means
Licence and copyright detection the checkout together with the author's unversioned working files beside it, 6810 files ScanCode Toolkit 32.5.0, SPDX licence list 3.27
Snippet and file match against public code, 2026-11-01 303 own files, 194678 lines: 107 modules, 93 files under tests/, 89 examples, 11 tools, doc/tclpdf.md, configure.ac, Makefile.in SCANOSS (scanoss-py 1.54.3, knowledge base 26.10 / 26.11.01)
Manual review of the risk areas font parsing, subsetting and shaping; PNG, JPEG and TIFF decoding; encryption; signing; XMP and PDF/A; generated lookup tables; build files reading, searches for foreign names and prefixes
Line comparison the modules and tools against the pdf4tcl fork 0.9.4.49 (214 files, 24610 file pairs), tcllib 1.21/2.0, jsPDF 4.2.1 with jsPDF-AutoTable 5.0.8, libhyphen 2.8.8 (hyphen.c) identical normalised lines, shared four-line windows, shared identifiers

The audit ran from 2026-10-29 to 2026-11-01, read-only, carried out by Claude Code agents on the author's behalf. Limits: SCANOSS compares normalised source text; a translation from C into Tcl cannot be recognised that way, so the scan rules out copied lines, not adopted logic. libjpeg, libtiff, HarfBuzz and tFPDF were not available locally for a text comparison. Where tclpdf follows them - decision rules in imageJpeg.tcl and imageTiff.tcl, the StemV estimate in font.tcl, the Arabic shaping order in forms.tcl - the source names them openly and SCANOSS reports no match in any of these files; whether those projects are in its knowledge base was not queried. pdf4tcl 0.8.1 was not compared. The editions of the standards the code was built against are named: ISO 32000-1:2008 and ISO 32000-2:2020 on the wiki page sources, ISO 19005-2 and -3 (one edition each, 2011 and 2012), ISO 14289-1:2014 and 14289-2:2024, ISO/TS 32005:2023 and WTPDF 1.0 in the author's working copies. The audit reports are working files of the author and not published; what the repository shows is the check-in that acted on them (section 4).

3. What was found

No foreign code in the package, no copyleft finding in versioned own code, and no finding rated suspicious. ScanCode found the author's MIT header in 303 of 327 own files and no foreign licence in them, only licence names in comments and the manual and one false positive: the header sentence "See the file license.terms for information on usage and redistribution" is the formula of the Tcl sources, so ScanCode reads it as the Tcl licence at score 55. The copyleft hits in versioned files are the AGPL option of the URW fonts (the option chosen is OFL 1.1) or false alarms; material under LGPL, GPL, MPL or proprietary terms lies only in unversioned directories that no check-in and no archive carries (measured on the 1.7 source archive). SCANOSS found no match in 299 of 303 files and none in the modules, tests, examples or manual. The four snippet matches, none a file match, are all in the build and test scaffolding:

File Lines Match Explanation
configure.ac 80-127 36 % of the configure of IngTcl 0.3 TEA template text; 32 of the 38 non-empty lines stand word for word in the TEA templates
tests/all.tcl 4-29 14 % of ooxml the author's own project (github.com/aschoepe/ooxml); the header is the TEA all.tcl
tools/archive.sh 7-204 94 % of ooxml 1.13.2 (Debian) the author's own build script
tools/mkversion.tcl 10-204 95 % of nacl the author's own project (github.com/aschoepe/nacl); see below

One value was taken over and is gone: the first version of encrypt.tcl (01724f5e6a, 2026-08-20) wrote /Length 256 into the encryption dictionary as the pdf4tcl fork does; review round 5 removed it as contrary to ISO 32000-2 Table 20 (7c8f62ca73, 2026-08-22). Besides that, the first tools/mkafm.tcl (e1795e6562) read pdf4tcl's glyph2uni.tcl - the Adobe Glyph List as data - at build time; 5c200fdea5 replaced it with Adobe's glyphlist.txt. One build tool is derived from foreign code and says so: tools/mkversion.tcl, which writes manifest.txt, is the author's port of Fossil's mkversion.c (BSD-2-Clause), dated 2024-2025 in its header and older than tclpdf; it is neither installed nor part of the binary archives. One line of encrypt.tcl, which sets a permission bit, is identical to a line of the fork; its form follows from the standard's one-based bit numbering and the code around it differs, so whether it was written independently cannot be shown either way - it stays unclear, with little weight. Everything else the comparisons share with other code is fixed by a standard or format: AES (FIPS 197), the SHA-2 constants (FIPS 180-4; 24 lines equal to the fork in the layout of the standard's table), the encryption algorithms and permission bits (ISO 32000-2), the names of the 14 standard fonts, xref lines and structure rules (ISO 32000-1), Paeth predictor and IHDR fields (PNG), CCITT codes (ITU-T T.4), Type 1 and CFF constants, OpenType field names, AGL glyph names, Unicode blocks, compositions and joining types, OIDs (RFC 3161, RFC 5652) and the Factur-X XMP schema, which accounts for the eight lines shared with jsPDF. tcllib shares only Tcl commonplaces and the TEA head of tests/all.tcl. hyphenate.tcl is an independent implementation of Liang's method: a flat dictionary instead of libhyphen's automaton, none of its extended pattern syntax, constants or names.

4. What was fixed

Check-in f144958ae7 (2026-10-29), released as 1.7.1 (0318611fc3, tag version-1.7.1), added eight notices:

Seven of these answer findings rated "to act on"; the eighth such finding concerned a stale tool path in the author's unversioned notes. The Permanent Marker line was the author's answer to a finding rated unclear.

A second round on 2026-11-01, check-ins 9a0e7abc29 and 9594fb8dbc, released as 1.7.2 (18f0fa3f95, tag version-1.7.2), added three more notices, each in the source archive only:

5. What stays open or is a matter of judgement

6. Numbers

Of the 61 findings, 35 are in order (five of them by the author's decision or statement: the URW licence choice, subsets without a name table, the FeRD samples, the specimen portrait, and the no-copying rule, which the author states was part of the commission of 2026-08-08 and which is in writing since 2026-08-29), 13 are fixed by the standard (the match is dictated by a standard or format), 2 are unclear (not decidable with the means at hand: one shared line in encrypt.tcl, and no text comparison against libjpeg, libtiff, HarfBuzz or pdf4tcl 0.8.1), 11 were to act on - eight acted on in f144958ae7 (1.7.1), three in 9a0e7abc29 and 9594fb8dbc (1.7.2), section 4 - and 0 are suspicious. How the package was made and which third-party material it contains is set out in the provenance note: https://fossil.sowaswie.de/tclpdf/technote/cb5371ca27