The goal of this project is to revive the historic Algol compiler written by Edsger Dijkstra in 1960 for the Electrologica X1 computer.

Sources downloaded from: https://gtoal.com/languages/algol60/x1algol/

Directories:

• compiler - sources of Algol-60 compiler in Pascal and C
• doc - documentation about X1 compiler, Electrologica X1 machine and Algol language
• examples - sample code to compile and run
• misc - helper stuff in progress
• scripts - disassemblers, decoders and other useful scripts
• simulator - implementation of a virtual X1-like machine
• tests - unit tests for compiler and simulator

Build

Compile the X1 Algol simulator from sources and install into /usr/local:

make
make install

Run tests:

make test

Expected output:

ctest --test-dir build/tests
...
        Start   1: arith.x1_to_ieee
  1/105 Test   #1: arith.x1_to_ieee ........................   Passed    0.01 sec
        Start   2: arith.ieee_to_x1
  2/105 Test   #2: arith.ieee_to_x1 ........................   Passed    0.01 sec
        Start   3: x1_machine.arith_add
...
105/105 Test #105: cli.lib_sum .............................   Passed    0.02 sec

100% tests passed, 0 tests failed out of 105

Total Test time (real) =   1.39 sec

Examples

A few demos are available in the examples directory:

$ cd examples

$ x1sim hello.a60
Hello `Algol'

$ x1sim man_or_boy.a60
-67

See full list on page examples/README.md.

Known bugs

Misplaced parentheses in the Pascal version of the compiler

The line 1415 arrc:= arrc - (arrd * store[tlsc-1]) mod d26; as per the Kruseman Aretz paper, must be arrc:= (arrc - arrd * store[tlsc-1]) mod d26; to ensure correct computation of storage functions of multi-dimensional arrays.

This is a Pascal transcription artifact.

Mistakenly disabled generation of passing of formals descriptors

Formal parameters of procedures directly passed as actual parameters to other procedures produce thunks (implicit subroutines) unnecessarily.

The lines

1074     then if (store[tlsc-2] = rlsc)
1075         and (fflag = 0) and (jflag = 0) and (nflag = 1)
1076         then begin if nid > nlscop
1077             then begin if (pflag = 1) and (fflag = 0)

Have two checks for fflag = 0 in nested if conditions. One of the checks is redundant. Removing the first one, in line 1075, restores the ability to generate descriptors allowing a more efficient parameter passing. This saves about 15% of stack depth for the "Man or boy" test.

It is unclear if this is a Pascal transcription artifact or a genuine performance regression in the original compiler.

Passing predefined functions as procedure arguments sometimes results in bad code

A numerical actual parameter of a procedure call immediately following a name of a predefined function, e.g. sin or cos, results in garbage object code. The check nid > nlscop in line 1076 supra fails because the variable nid still holds the ID of a predefined function when the next parameter is not an expression and does not contain an identifier, resulting in processing the number as if it were a predefined function.

A fix is to add a flag which is set when such a name is encountered first (line 1104), and is used to trigger resetting nid to an arbitrary number greater than the predefined function boundary (nlscop) at the end of production_of_object_program.