Heads up: this is a technical post about porting an LDMud mudlib across 14 years of driver evolution. If you usually come here for game design or play notes, you can safely skip this one.

We run a small Spanish MUD called Endor that was built on LDMud 3.3.720, released in 2011. The mudlib had been running unchanged for years on a legacy host. When we decided to move it into our Docker cluster, we hit a wall: LDMud 3.3.720 has known 64-bit bugs that cause the driver to segfault on player login. The only way forward was to port the mudlib to the current stable version, LDMud 3.6.8.

This post documents every change we had to make.

Why not just recompile 3.3.720?

We tried. LDMud 3.3.720 compiles cleanly on Debian 12 with gcc 12 using -O0 (no optimization). The driver starts, loads the mudlib, and reports “LDMud ready for users.” But the moment a player tries to log in, the driver crashes with a segfault (exit code 139).

This is caused by known 64-bit bugs in the 3.3.x branch — issues with remove_interactive, memory allocation on x86_64, and time_t overflow that were fixed in later versions but never backported to 3.3.720.

We also tried:

  • Building with -fno-strict-aliasing -fwrapv and other UB-prevention flags — still crashes.
  • Building with gcc-10 (Debian bullseye) — still crashes. It’s a code bug, not a compiler issue.
  • Building as 32-bit (-m32) — fails to compile because sprintf.c has 64-bit-only code paths.

None of these worked, so we decided to port the mudlib to 3.6.8.

1. Character encoding: ISO-8859-1 to UTF-8

LDMud 3.6.8 enforces UTF-8 encoding for all LPC source files at the compiler level. There is no flag to disable this.

Endor is a Spanish MUD. About 7,200 source files were encoded in ISO-8859-1, full of accented characters in room descriptions, NPC dialogue, and item names. We wrote a Python script that detects files with high bytes that aren’t valid UTF-8 and converts them:

data = open(path, 'rb').read()
if not any(b > 127 for b in data):
    continue  # pure ASCII, skip
try:
    data.decode('utf-8')
    continue  # already valid UTF-8
except:
    text = data.decode('iso-8859-1')
    open(path, 'w', encoding='utf-8').write(text)

Save files (.o), text files, and other data files were converted too.

2. Line endings: CRLF to LF

Many files had Windows-style \r\n line endings, visible as <0x0d> in git diffs. We converted them to Unix \n line endings across ~10,600 files with perl -pi -e 's/\r\n/\n/g; s/\r/\n/g' to keep the repository clean and avoid potential issues with the LPC compiler.

3. System headers from 3.6.8

The mudlib’s sys/ directory needed the header files from LDMud 3.6.8 (e.g. configuration.h, object_info.h, interactive_info.h, driver_info.h, charset.h). These define the constants used by the compatibility functions described below.

We copied all sys/*.h files from the LDMud 3.6.8 source tree into the mudlib.

4. Deprecated efuns in the master object

The master object is loaded before the simul_efun object, so any deprecated efuns called by the master must be replaced directly with their new equivalents:

Old efun Replacement Header needed
seteuid(uid) configure_object(this_object(), OC_EUID, uid) configuration.h
query_once_interactive(ob) efun::object_info(ob, OI_ONCE_INTERACTIVE) object_info.h
query_idle(ob) efun::interactive_info(ob, II_IDLE) interactive_info.h
query_shadowing(ob) efun::object_info(ob, OI_SHADOW_PREV) object_info.h
shadow(ob, 0) efun::object_info(ob, OI_SHADOW_NEXT) object_info.h

5. Master function signature changes

LDMud 3.6.8 changed the expected signatures of several master applies:

  • log_error: from (string file, string err) to (string file, string err, int warn, int line)
  • runtime_error: from 4 arguments to 6 — added mixed culprit and int caught
  • make_path_absolute: from varargs (string path, string user) to just (string str)

These had to be updated in both the header declarations and the implementations.

6. Compatibility simul_efun header

LDMud provides official replacement simul_efuns for every deprecated efun. We created a single secure/compat_368.h file that #includes at the top of simul_efun.c, containing all the replacements the mudlib needs:

seteuid, query_once_interactive, query_idle, query_shadowing, query_snoop, query_mud_port, query_ip_name, query_ip_number, query_load_average, set_is_wizard, debug_info, assoc, order_alist, insert_alist, shadow (2-arg compat), enable_commands, set_heart_beat, set_prompt, tail, hash, and many more.

Functions that the mudlib’s own simul_efun.c already implemented (like cat, strlen, file_name, filter_array, etc.) were skipped to avoid redeclaration errors.

One subtlety: debug_info.h and driver_info.h have overlapping #define constants (TRACE_TYPE, TRACE_NAME, etc.). We had to #undef the conflicting symbols before including debug_info.h.

7. set_this_object restriction

LDMud 3.6.8 does not allow calling simul_efuns while set_this_object() is in effect. The mudlib’s cat() function used set_this_object(previous_object()) and then called strlen(), which is now a simul_efun. The fix was to replace strlen() calls inside cat() with efun::sizeof() directly.

8. Multiple visibility modifiers

LDMud 3.6.8 disallows private static or static private — only one visibility modifier is allowed per declaration. Since private already implies static in LDMud 3.5+, a bulk replacement across ~89 files was sufficient:

private static → private
static private → private

9. Non-ASCII characters in identifiers

LDMud 3.6.8 does not allow non-ASCII characters in identifier names (macro names, variable names, function names). Our mudlib had a few #define names with Ñ and ó:

  • HAB_LEÑA → HAB_LENYA
  • PUÑOFERREO → PUNYOFERREO
  • NIV_LEÑA_AVENTUREROS → NIV_LENYA_AVENTUREROS

These were renamed throughout the mudlib.

10. Type checking fixes

LDMud 3.6.8’s type checker is significantly stricter than 3.3’s. Several files needed individual fixes:

  • Variables declared as mapping but used as object for call_other — changed to mixed with explicit (object) casts where needed.
  • Variables declared as int but receiving mixed values from arrays — changed to mixed.
  • Function signatures where a child class redeclared a function with different varargs or argument count — fixed to match the parent.
  • Forward declarations in header files that didn’t match the implementation signature.

Each file was fixed individually rather than disabling type checking globally.

11. Master object must be pure 7-bit ASCII

This was the trickiest encoding issue. LDMud 3.6.8 uses iconv to convert source files from their declared encoding to UTF-8 internally. The encoding is determined by the H_FILE_ENCODING driver hook. But the master object is compiled before any LPC code runs, so the hook cannot be set in time.

Without the hook, the driver defaults to 7-bit ASCII — and iconv("ascii" → "utf-8") in glibc strictly rejects any byte above 127. This means every file in the master’s #include chain must be pure ASCII, even if the rest of the mudlib is UTF-8.

The official migration document states:

Because the master must be loaded first to set the encoding for other files using the file_encoding(H) hook, the master itself must be written in 7-bit ASCII only. If you need Unicode characters you can use Unicode escape sequences like \u2165 or \U0001F603.

We had non-ASCII characters in both comments and string literals across the master and all its included files (about 13 files total). The fix:

  • Comments: replaced accented characters with their ASCII equivalents (inicialización → inicializacion).
  • Active string literals: replaced non-ASCII characters with \uXXXX escape sequences ("¿Qué?\n" → "\u00bfQu\u00e9?\n", "empuñar" → "empu\u00f1ar").
  • Commented-out code: also used \uXXXX escapes so the strings would work correctly if uncommented.

Then, in inaugurate_master(), we set the hook for every subsequent file:

set_driver_hook(H_FILE_ENCODING, "UTF-8");

This way the master compiles as ASCII, and every other file in the mudlib (simul_efun, login, game objects, etc.) loads as UTF-8.

Note: we initially discovered this on x86_64 (production) while it worked fine on arm64 (local Mac). The difference is that macOS’s iconv is lenient with ASCII-to-UTF-8 conversion and passes through high bytes, while glibc on Linux is strict and rejects them. The behavior on Linux is correct per the spec.

12. Interactive encoding for modern clients

By default, LDMud 3.6.8 assumes player connections use ISO-8859-1 encoding. Modern MUD clients (Atlantis, Mudlet, or browser-based clients) typically send UTF-8. If a player types an accented character like í, a UTF-8 client sends 0xC3 0xAD (two bytes). With the default ISO-8859-1 setting, the driver interprets each byte separately, producing two garbage characters instead of one — and commands with accents (like telepatía) fail to match.

The fix is to set the interactive encoding to UTF-8 as early as possible in the connection flow:

#include "/sys/configuration.h"

public nomask int logon() {
    configure_interactive(this_object(), IC_ENCODING, "UTF-8");
    // ... rest of logon
}

The result

Endor’s mudlib loads and runs on LDMud 3.6.8. Players can log in, the database connection works (via a shared MariaDB Unix socket), accented commands work from modern clients, and the driver is stable — no segfaults, no crashes.

The driver itself required zero patches. Every change was in the mudlib or in the compatibility simul_efun layer. LDMud’s official deprecated/ directory was invaluable — it provided drop-in replacements for every removed efun.

What we learned

  • Don’t try to keep the old driver alive. We spent hours trying to make 3.3.720 work with modern compilers before accepting that the 64-bit bugs are unfixable without driver patches. Porting the mudlib to 3.6.8 was the right call.
  • LDMud’s deprecated/ directory is excellent. Every removed efun has an official simul_efun replacement with the exact same behavior. Use them.
  • Convert encoding first. The UTF-8 requirement is absolute — there’s no flag to disable it. Do the bulk conversion before anything else.
  • The master must be pure ASCII. This is documented but easy to miss. If your mudlib is in a language other than English, you almost certainly have accented characters in the master’s string literals. Use \uXXXX escapes and set H_FILE_ENCODING to "UTF-8" in inaugurate_master().
  • Set interactive encoding to UTF-8. The default ISO-8859-1 breaks accented input from modern clients. A single configure_interactive call in the login object fixes it.