Differences from vamos¶

volamos is a spiritual successor to vamos, the Python implementation of the same idea, and the two are compared against each other continuously — tools/compare_vamos.py and tools/compare_amitools_suite.py run the same binaries through both and fail on any unexplained disagreement.

Most disagreements that surface turn out to be volamos bugs, and get fixed. This page documents the ones that went the other way: cases where the two runtimes genuinely differ and volamos's behaviour has been verified against real Amiga hardware.

These are recorded rather than "fixed" because there is nothing to fix on volamos's side. If you are porting a test or a workflow from vamos and hit one of these, this page is the explanation.

How these were verified¶

Every entry below was checked against a real Kickstart running under Copperline, booting a real Workbench 3.1.4 filesystem, with the guest's own Shell redirecting the program's output to a host-visible file — so the comparison is against actual dos.library/utility.library/mathieeedoubbas.library ROM code, not against either runtime's idea of what the ROM does.

That matters more than it might sound. The NDK autodocs describe intent, and at least one entry below is a case where the documentation describes behaviour real AmigaOS does not implement — so documentation alone was not treated as sufficient evidence.

tools/compare_three_way.py automates this three-way comparison (volamos / vamos / real Kickstart). Note it needs a --model matching the --rom you give it; a mismatched pair boots nothing and silently produces an empty real-Kickstart column.

The differences¶

`WriteChars()` and a trailing NUL¶

char msg[] = "Hello, world!?\n";
WriteChars(msg, sizeof(msg));      /* sizeof includes the implicit NUL */

WriteChars is a raw byte-count write. Real dos.library writes exactly the count it is given, so the trailing NUL that sizeof legitimately includes is written:

00000000: 4865 6c6c 6f2c 2077 6f72 6c64 213f 0a48  Hello, world!?.H
00000010: 656c 6c6f 2c20 776f 726c 6421 3f0a 00    ello, world!?..

volamos writes all 16 bytes, matching hardware. vamos drops the NUL.

`CheckDate()` does not validate `wday`¶

utility.library/CheckDate's own autodoc carries this under BUGS:

The wday field of the ClockData structure is not checked.

So a ClockData with wday = 7 — outside the valid 0–6 range — is still accepted as a valid date. volamos matches that (and real Kickstart 3.1 confirms it); vamos rejects the date.

This is the entry worth remembering when weighing documentation against hardware: here the documentation documents a bug, and the bug is the real behaviour.

Sign of a quiet NaN from a math domain error¶

For results that are mathematically undefined — 0.0/0.0, acos/asin outside [-1,1], log of a negative number — both runtimes return a quiet NaN, but with opposite sign bits. Real mathieeedoubbas.library's IEEEDPDiv(0.0, 0.0) returns a positive-signed NaN regardless of input sign, which is what volamos returns; vamos returns a negative-signed one.

The NaN payload bits below the sign differ between real hardware and volamos, and that is treated as implementation-specific detail, in the same category as ordinary transcendental-function rounding noise.

The command line carries a trailing space¶

When a launched program has at least one argument, real AmigaOS's command-line buffer carries a trailing space before the final newline — "foo bar baz \n", not "foo bar baz\n". Confirmed against real Kickstart 2.0, 3.0 and 3.1 hardware.

volamos reproduces this; vamos does not, so any test asserting on the exact command-line bytes will differ by that one space.

Kickstart 3.2 alone drops the trailing space — a real, intentional AmigaOS version difference rather than a bug. volamos targets 3.1 first and follows 3.1 here.

`APF_DirChanged` during a `MatchFirst`/`MatchNext` scan¶

dos/dosasl.h defines the flag as "ap_Current->an_Lock changed since last MatchNext call". Real dos.library sets it on the first entry reported after descending into a directory, clears it for subsequent entries in that same directory, and sets it again when leaving.

volamos implements exactly that. vamos never sets the flag, so a program that drives its behaviour off directory transitions — the real List command uses it to decide when to print a new Directory "..." header — has nothing to work from.

What is not on this page¶

Differences where volamos is the one that is wrong get fixed rather than documented, so they do not appear here. Some are still open as bugs — see the issue tracker — and a few known ones are listed as visible failures in the comparison harnesses' own tables rather than hidden.

vamos also does not read volamos's .uaem sidecar metadata (protection bits, comments, timestamps), so any comparison whose output depends on dates or protection bits is not meaningful between the two runtimes and is scoped out of the harnesses rather than tracked here.