Lisp Machine Museum Documentation Examiner
Genera Museum Documentation
Knowledge Base›MIT CADR›MIT CADR font sources and recovery

MIT CADR font sources and recovery

Conclusion

The source-represented CADR font corpus can be recovered directly from the public source tree. There is no reason to mine a load band, emulator heap, or Symbolics VLOD for these fonts. The repository's extract-cadr-fonts.py reconstructs the archived PDP-10 words, decodes the font source formats, and emits portable BDF fonts and PNG font sheets.

Against revision 8e978d7d1704096a63edd4386a3b8326a2e584af of the public src/lmfont tree, the default extraction emits 150 font artifacts representing 88 logical names. The difference is intentional: 62 source variants have materially different glyphs or metrics and are preserved separately rather than silently collapsed.

The later dedicated public CADR Fonts pipeline supersedes this museum-local 150-artifact build for public installation and release. It corrects the AST raster-height model for BUG: the 32-row authored AST form and 33-row BUG-KST form are retained as distinct artifacts, giving 151 source artifacts. It also keeps a separate inertly decoded 49-QFASL runtime profile and publishes source/runtime Unicode BDF and OTB packages. The source and resident profiles must not be collapsed when selecting a font for an exact interface target.

These outputs are normalized preservation derivatives. They retain character codes, declared metrics, and every recoverable set pixel, including pixels outside some Alto records' declared extents. They are not claimed to be byte-identical QFASL files, runnable CADR font objects, or byte-for-byte reproductions of what historical FNTCNV would load.

Provenance and licensing

The input is the public mietek/mit-cadr-system-software repository at commit 8e978d7d1704096a63edd4386a3b8326a2e584af. Its src/README identifies the material as an early MIT AI Laboratory CADR snapshot, System 46, recovered from four ITS backup tapes. The source license is the three-clause BSD license, despite the README's informal description of it as an "MIT license". The exact src/LICENSE is copied beside the generated assets as LICENSE.source.

There is an unresolved date discrepancy in the public provenance record. The tape labels transcribed in src/README say 10/17/81, while the extracted archive filename contains 101780. This page does not choose between them; establishing the exact tape date remains a TODO.

Why the files need decoding

Files under src/lmfont are not ordinary host-byte copies of their historical ITS contents. They use Alan Bawden's evacuated representation of PDP-10 data. The first stage therefore reverses the text escapes and quoted binary-word encoding to recover 36-bit words. The implementation follows the public itstar unpacker rather than treating the apparent ASCII as a modern text file.

The extractor then handles these source containers and formats:

Input Recovery rule Evidence
arc.ast's Read the ARC!!! directory, allocation descriptors, and block chains; ignore directory entries marked deleted; parse each live member as AST font source. DMARCD describes ARC!!! as its "new flavor" archive convention. The archive logic is checked against the contemporary DMARCD implementation.
ar1.1 Read a second ARC!!! container whose members are Xerox Alto CONVERT font sources. Preserve both ITS names, select the greatest numeric second name, and retain character-pointer and declared-extent findings for every member. The public AR1 1 archive is decoded with the same ARC logic; its payload structure matches the CADR Alto reader and the contemporary Xerox format description.
.ast Read the header and form-feed-delimited glyph pages, including per-glyph width and left-kern metrics. The recovered structure follows the CADR compiler's RD-AST reader.
.kst Read the PDP-10 interchange header, character records, raster words, and end marker. CADR FNTCNV identifies KST as the PDP-10 interchange format; the ITS KST format document supplies the record layout.
.al Decode Xerox Alto CONVERT font data using the semantics implemented by CADR FNTCNV. See the ALTO .AL loader and the Xerox font-format description, pages 12–13.
.cldfnt Optionally decode the compiler's textual cold-font dump. MAKE-COLD-FONT writes this representation. It is excluded by default because the archive already contains TVFONT's AST font source.

An ARC allocation descriptor counts logical 1024-word file blocks; that count is not the number of physical nodes in the data-block chain. The extractor derives the declared word length from the logical count, then independently checks every physical node's header/trailer agreement, predecessor and successor links, final marker, and aggregate word length. Both counts are retained in the catalog.

Compiled .qfasl, .oqfasl, and .unfasl files in the directory are deliberately not inputs to this source-first process.

Extraction result

The inspected arc.ast's directory has 72 entries: 71 live AST font source members and one ignored older CLAR12 entry. All 71 live members parse. The separate ar1.1 archive has 63 live Alto members under 62 logical names because PRONTO has versions 1 and 2; numeric ITS version selection chooses version 2. After source precedence and duplicate handling, the emitted set comprises:

  • 71 fonts from live AST font source archive members;
  • 11 additional fonts from standalone AST font sources;
  • six emitted KST sources, including four new logical names and two divergent variants;
  • 62 selected Alto sources from ar1.1, adding the logical names CLRE14 and PRONTO and retaining 60 materially different historical variants.

Twelve alternate sources normalize to the same metrics and glyphs as an already emitted representation and are recorded in the catalog without being emitted again: standalone AST font sources for 43VXMS, APL14, ARR10, and TONTO; KST sources for BUG and TOG; and standalone Alto sources for TR10, TR10B, TR12, TR12I, TR14, and TR18 that equal their ar1.1 members. "Same" here means semantic equality after decoding, not byte-for-byte equality of unlike containers.

The 62 emitted variants consist of two KST variants and 60 ar1.1 Alto variants:

  • ARROW-KST and BIGFNT-KST retain the divergent KST representations;
  • names ending in -AL-AR1 retain historical Alto representations when a higher-precedence AST or KST representation has the same logical name.

The archive catalog also retains the unselected PRONTO 1 source record. It is not emitted as another font because contemporary numeric version selection chooses PRONTO 2.

Character-pointer partial Alto members

All 63 ar1.1 members have a valid archive and Alto structure, but seven member versions contain one or two objectively impossible control-character pointers. The extractor does not mask, reinterpret, or invent those glyphs. It omits only the bad character record and marks the character-pointer integrity as partial. Six selected outputs are affected:

Selected source Omitted octal codes
CLARGK 1 003, 004
HIP10A 1 003, 004
PRNT10 1 001
PRONTO 2 003
PRT12B 1 001
SMT14A 1 003, 004

The seventh partial member is the unselected PRONTO 1. Its version-2 successor repairs code 004 but retains the invalid code 003. --strict still succeeds because these are explicit partial recoveries rather than rejected sources; the catalog records each bad raw pointer, error, omitted code, member hash, and selection decision.

Declared-extent recovery in Alto members

Character-pointer integrity and raster-extent recovery are separate findings. A member can have a complete pointer table while containing set pixels outside the dimensions declared by its Alto metrics. In this source revision, 45 emitted ar1.1 fonts have 142 character-code records with horizontal set pixels beyond the declared advance. In the selected PRT12B 1 member, code 030 has a second nonzero row at Y=15 even though the declared line height is 15, whose zero-based nominal range ends at Y=14.

The extractor preserves those observed pixels by widening or extending the normalized glyph raster as needed while retaining the declared advance and line height as metrics. This is an explicit recovery transformation, not a correction to the historical source and not evidence that the affected records were byte-for-byte loadable by the historical FNTCNV path. The catalog reports declared-extent recovery separately from bad character pointers so that a structurally complete member is not mislabeled as pointer-partial. Whether any out-of-extent pixels were intentional, tolerated by a particular historical build, or reflect source damage is TODO; the decoded records alone do not establish that explanation.

The public source files tr12.al and tr14.al are byte-identical. The catalog records that archival fact without inferring why the two names were retained.

Reproduce the tracked assets

From this repository's root, set CADR_CHECKOUT to a checkout of the public source repository and verify its revision before extracting:

CADR_CHECKOUT=../mit-cadr-system-software
test "$(git -C "$CADR_CHECKOUT" rev-parse HEAD)" = \
  8e978d7d1704096a63edd4386a3b8326a2e584af
python3 scripts/extract-cadr-fonts.py \
  "$CADR_CHECKOUT/src/lmfont" \
  --output docs/assets/mit-cadr-fonts \
  --clean \
  --omit-json \
  --strict

--clean replaces only the extractor-owned asset paths and refuses a destination containing unrecognized entries. This makes a regeneration remove obsolete filenames instead of silently retaining artifacts from an older source or naming rule.

The tracked derivatives are:

  • catalog.json, containing source hashes, decoded metrics, archive observations, duplicate decisions, variants, and output paths;
  • font-usage-catalog.json, the separately maintained evidence audit for every logical name. It is not generated from glyph data because a source filename or glyph shape is not evidence of purpose;
  • bdf/, containing 150 portable bitmap fonts;
  • sheets/, containing 150 PNG font sheets whose glyph labels use octal character codes;
  • LICENSE.source, the license shipped with the public source snapshot.

Omit --omit-json when per-font normalized JSON is useful for analysis. The catalog, BDF, and sheets are still written either way.

Source-only coverage boundary

The 88 logical names are all names recoverable from the AST, KST, Alto, and optional cold-font source representations in this exact snapshot. They are not every compiled font present in src/lmfont, and this page does not claim that every CADR font ever used has been recovered.

Nineteen additional .qfasl files decode as compiled FONT objects. Seventeen have no matching source representation in this snapshot and add these runtime logical names: 20VR, 31VR, 40VR, BIGVG, CPT-13FG, CPT-HL10, CPT-HL10B, CPT-TR10I, GERM35, HL12BI, MEDFNB, S30CHS, S35GER, SAIL12, SEARCH, SHIP, and TR12B1. The remaining two compiled files, N43XMS and NTOG, are older compiled versions of the source-backed 43VXMS and TOG names.

They are now recovered by the deliberately separate compiled QFASL font pipeline. That extractor emits the serialized runtime objects as BDF, normalized JSON, and PNG derivatives while preserving their compiled-artifact provenance. It does not claim to reconstruct the missing authoring source or add these objects to the 88-name source-backed corpus.

Usage audit

Font names and glyph appearance are not evidence of purpose. The complete font usage audit searches the pinned source, manuals, bug reports, load declarations, editor metadata, and binary document output for every logical name.

It establishes direct executable roles for 15 of the 88 source-backed names, qualified printer/document roles for three, and contemporary documented roles for three. The other 67 remain explicit TODOs because the snapshot establishes only reported use without a role, standard loading, source compilation, or source survival. The audit also treats all 17 QFASL-only names separately: System 46 source establishes roles for MEDFNB and SHIP, while maintained LM-3 System 303 source establishes the cross-version role of SEARCH with an explicit non-identity caveat.

This separation matters because the tracked 150 files include 62 alternate source representations. Usage belongs to a logical name unless primary evidence identifies a particular AST, KST, or Alto variant.

Open questions

  • Resolve the 10/17/81 versus 101780 source-tape date discrepancy from primary media documentation.
  • Locate additional primary evidence for the 67 source-backed and 14 compiled-only logical names whose purpose remains TODO after the complete literal-reference audit.
  • Compare both the source-derived and compiled-artifact outputs with fonts loaded by a running System 46 image. That would test the complete historical compilation and loading path, but it is not required to recover the public glyph data.

Last verified: 2026-07-26.

Try “search window system”, “open genera”, or “help”.