parseArguments() split its argument string with a std::istringstream.
That single use pulls in libstdc++'s iostreams and locale support, with
all the standard facets (money_get, time_get, their wchar_t variants...)
and their exception tables: a plain "hello world" console application
grows from about 210K to about 1.2M.
That is more than the 1M partition the default SIZE resource in
Retro68APPL.r asks for. Under System 7, loading a code segment then fails
with memFullErr, the assert in Retro68LoadSegment calls DebugStr, and
without a debugger installed the application stops at launch with an
"unimplemented trap" system error. System 6 without MultiFinder ignores
the SIZE resource, which hides the problem there.
Split the string on ';' with std::string::find instead. The result is
the same: empty fields still count as 1, and there are at least two
numbers.
Pick up upstream newlib commit 588c19ecdbcca1e70030688c1b08c342ada01b84 ("Fix memcpy alignment problems for m68k architecture", Yuichi Nakamura), which the bundled newlib 4.6.0 predates.
Under MISALIGNED_OK=0 (68000) an unaligned copy longer than 64 KiB was truncated, because the residue loop used "dbra d1" over the whole length. Also fixes 68020 misaligned access, word-aligned 68000 access, and the signed loop comparison that broke copies >2 GiB.
Fixes#315
DoDriverIO's types, and NumVersion is a plain integer there. Recognize when
the Universal Interfaces are present by the UNIVERSAL_INTERFACES_VERSION rather than by
the header's presence, and with Multiversal declare only what it lacks.
Builds an 'ndrv' the Driver Loader loads from the Extensions folder:
links with ndrv.lds and an export list, runs MakePEF, and Rezzes the PEF
with RetroPPCndrv.r into <name>.bin and <name>.dsk.
KIND GENERIC, AIM or SIM writes the export list for a Device Manager
driver (DoDriverIO), an ATA Interface Module (ThePluginDispatchTable) or a
SCSI Interface Module (LoadSIM), each with TheDriverDescription. EXPORTS
takes a list file instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
One member, kDropInAdditionCFrag, no 'SIZE'. Mac OS skips a driver file
without a 'cfrg' in its resource fork.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Lays out GCC's XCOFF sections the way MakePEF expects for an 'ndrv':
read-only data goes in .data, since a driver is never memory-mapped, and
.data starts with a pad word against a ROM DisposePtr bug. From
elliotnunn/classicvirtio, MIT-licensed. Installed with the .cmake files.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Mini vMac launcher builds its boot disk from scratch and copies only
the System file, the debugger and a fixed handful of extensions from the
configured system image. Anything else a program under test needs is
absent -- notably the 'AppleTalk' file, which carries the .XPP and .DSP
drivers on System 6/7, so OpenDriver(".DSP") returns fnfErr on every
LaunchAPPL boot even though the system image has it.
Add a repeatable --system-extra-file NAME option (config-file key
system-extra-file). Each named file is copied by name from the system
disk's blessed folder when present, right after the debugger file, using
the existing CopySystemFile path; a name missing from the system disk is
skipped, so one config can serve images that differ. Documented in
LaunchAPPL.cfg.example with the AppleTalk file as the example.
Verified on macOS with a System 6.0.8 image: an app that opens .DSP
fails without the option and succeeds with `--system-extra-file
AppleTalk`.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KKngnYbKgJyQ78EqPQZAGL
The previous `_realloc_r` implementation would leak the original memory block (`ptr`) when the in-place `SetPtrSize` call failed and a new block had to be allocated.
This change also adds correct handling for `realloc(ptr, 0)`, which should free the block and return `NULL`.