Software Architecture, Design & Engineering by Nicholas J. Kisseberth.
Build PDP-11 programs on a modern host — preprocess, compile, optimize, assemble, link — into real 2BSD a.out binaries, and run them unmodified under a user-mode PDP-11 UNIX simulator, faithfully across every era from the First Edition and Sixth and Seventh Editions through 2.11BSD, from one set of tools.
The same tools build for the First Edition through 2.11BSD. A universe — chosen with --universe or PDP11_UNIVERSE — selects that era's headers, C library and system-call personality; the objects and archives stay one a.out family, so everything links together and the simulator runs the result.
The assembler's own operand tables define every instruction's operand count, access class and data type. The disassembler and the simulator read the same tables, so all three agree about the machine by construction rather than by care.
The PDP-11 FP11 (and the FIS instructions) are evaluated bit-exactly in software. Results do not depend on the host FPU, so the same program produces the same bits on every machine that can build the tools.
The tools
The compiler driver
Runs cpp → c0 → c1 [→ c2 with -O] → as → ld in order and propagates the chosen universe to each pass. Ritchie's own cc, made cross and universe-aware.
The C preprocessor
The pristine 2.9BSD cpp (Reiser lineage) — the one the PDP-11 cc was built against, and where an era's #ifdefs are dated.
cc pass 1 — the parser
Ritchie cc's front end: reads preprocessed C and writes the intermediate form the code generator consumes.
cc pass 2 — code generator
Ritchie cc's PDP-11 code generator: turns c0's intermediate form into assembly.
The peephole optimizer
Ritchie cc's object-code improver, run over c1's assembly under -O.
The assembler
A C reimplementation of the 2BSD as, with historical --isa and --aout axes. It owns the operand tables das and apsim read, so all three agree about the machine.
The link editor
The authentic 2.9BSD ld on an LP64 host: resolves, lays out and relocates into the era's native a.out — including the 0430/0431 auto-overlay formats.
Extract string literals
Bill Joy's 1978 xstr: pulls a C program's string literals into one shared strings file and hashes them.
One universal C library
A single libc.a and crt0 that serve every era: the universe is chosen at link (ld stamps __univ), crt0 records it, and the routines whose ABI moved branch on it at run time. Plus curses and termlib.
List symbols
The symbol table of a.out objects, executables and archives.
Segment sizes
Text, data and bss sizes, and their sum, of a.out files.
The disassembler
The inverse of as, covering every era the assembler does; it splits a symboled a.out into per-object listings and goes linear on a stripped one.
Remove symbols
Removes the symbol table and relocation from an a.out — the 2.9BSD strip.
Copy / extract
Copies an a.out, or extracts its loadable image (-O binary) or a single text/data segment (-j) — the a.out-only analogue of objcopy.
The archiver
Authentic 2BSD ar, producing archives byte-identical to the originals.
The archive index
Builds the __.SYMDEF table of contents in an archive, so ld resolves it in one pass.
The PDP-11 UNIX simulator
Interprets PDP-11 instructions and translates the guest's sys traps into host ones, in one process. Runs unmodified binaries across the whole lineage — First Edition through 2.11BSD — with auto-overlays, job control, sockets, a bit-exact FP11 softfloat and a real adb debug channel.
Documentation
What the toolchain is, the one-idea universe model, the tools by role, and how to build and run.
OpenHow the merged toolchain is put together: the universe mechanism, the shared a.out object model, the pipeline, and faithfulness as the correctness standard.
OpenBuild the toolchain, pick a universe, compile and run PDP-11 programs, and drive the pipeline by hand.
OpenA category-by-category comparison of pdp11-apsim's system-call coverage against the reference vax11-apsim, and the calls both deliberately refuse.
Open