v2.19pre2

Bugfixes:

• format mz: stack segment was never written - a fault if it isn't
  located at the end of the binary; see stack1.asm.
• format mz,pe: start offset for first segment may have resulted in
  an invalid binary; see org1.asm & org2.asm.
• fastcall 32-bit (MS style): registers may have hold float arguments.
• no 'implicit far jmp/call to near label' error in pass one; see
  group7.asm.
• option -D: accept white spaces if option is enclosed in double quotes
  ( Masm compatible ).
• subtraction of 2 labels may have caused error "constant value too large";
  see minus1.asm.
• if .model wasn't flat, but segments with flat attribute existed, they
  may not have been listed in symbol table.
• regression in v2.05-v2.18, DUP operator: if an undefined symbol was
  contained in first operand, error message may have been omitted; see
  struct43.asm.
• local labels were accepted by PUBLIC directive without -Zm being set;
  see public3.asm.
• omf format: incorrect special handling of assembly-time variable fixups
  removed; see label9.asm.

Other changes:

• write listing reworked.
• option -pe: add segments with combine type stack - they may contain
  initialized data.
• JWasmr.exe: option -Sz removed.
• PUSHW will emit warning "magnitude of offset exceeds 16-bit" if
  operand is a 32-bit offset.
• using 'option oldmacros' will cause a warning only.
• option -pe: info section ".drectve" will be scanned and some linker
  options are recognized and handled.
• PUBLIC directive extended to allow setting the export attribute.
• PAGE directive: syntax check added.
• .model flat: ensure variables in 16-bit segments can be accessed
  directly if segment register is assumed flat ( Masm compatible );
  see flat2.asm.

v2.18

Bugfixes:

• omffixup.c, debug version only: a superfluous %s in a debug display
  may have caused an exception for output format omf.
• invoke directive: vararg parameter now rejected if used as argument;
  see vararg1.asm.
• don't add "assume cs:dgroup" in tiny model if format is coff/elf or
  model is flat; see tiny3.asm.
• default fastcall for 32-bit (MS style) caused a GPF for non-proc
  symbols; see fastcms.asm.
• listing: when segment directive created a list entry, the offset size
  may have been wrong.
• omf format: externdef may have caused a PUBDEF record to be written
  in pass 2, resulting in an invalid object module.
• invoke directive: if cpu was 8086, and the argument was a constant 0,
  a wrong value may have been pushed; see invoke56.asm.
• jwasm may have entered an infinite loop if a structured variable wasn't
  defined (yet), but a member name matched a local name; see undef2.asm.

Other changes:

• output format -djgpp: support added for 16-bit offset fixups.
• in listings, for directives, offset of 16-bit segments displayed with
  width 4.
• OWDOS32.mak added. Creates a 32-bit version that uses the native Open
  Watcom runtime for extended DOS. Open Watcom 2.0 is recommended to
  create this variant.
• listing: relocations in code marked with lowercase letters.

v2.17

Bugfixes:

• option renamekeyword: if the keyword was renamed multiple times it
  did work only if the first occurance of this option was after an
  instruction or directive that triggered "fastpass"; see renkeyw.asm.
• option renamekeyword: keywords weren't restored to their default at
  end of pass; see renkeyw2.asm.
• option renamekeyword: renaming an already renamed keyword may have
  caused an infinite loop; see renkeyw3.asm.
• size operand of DUP operator was checked to be positive, but wasn't
  checked for 32-bit overflow; see constnt3.asm.
• direct memory access with flat override ( for constant addresses )
  may have generated error "cannot access label through segment registers";
  see flatgrp3.asm.
• regression in v2.16: invoke may have generated a "PUSHD" in 64-bit code
  if .model wasn't flat; see invoke54.asm.
• FLAT: override may have suppressed address size prefix; see
  flatgrp4.asm.
• invoke, 64-bit, non-win64, vararg: 32-bit registers were rejected as
  arguments; see invoke55.asm.
• display of assembly-time was correct only if CLOCKS_PER_SEC was 1000.
• listing of embedded structs/unions: the printed start offset was wrong
  if the struct/union wasn't located before any code/data lines; see
  struct42.asm.
• djgpp output format was broken.
• COMM directive: if a type was given instead of a size, the type wasn't
  stored in the symbol and hence couldn't be used in expressions; see
  comm7.asm.

Other changes:

• option stackbase: may be used in conjunction with "assume ss" to
  define a PROC where CS and SS differ in offset magnitude ( 16-bit
  code with 32-bit stack and vice versa ); see stkbase1.asm, stkbase2.asm.
• format -pe: 64-bit segments defined with USE64 are added to internal
  "flat" group if model is flat. Allows to mix 32- and 64-bit segments in
  .model FLAT; see flatgrp5.asm & flatgrp6.asm.
• simplified segment directive .code with name argument in tiny model
  always added the segment to dgroup. Now this is done only if segment's
  offset size matches the model's offset size; see tiny2.asm.
• RETs inside PROCs will now show up in listing file like in Masm; see
  lst3.asm.
• in listings, offset of 16-bit segments now displayed with width 4.
• offsets/displacements that are too large will now always cause error
  "constant value too large"; see offset17/18/19.asm.
• output format -djgpp ( special coff variant ), previously optional only,
  is now activated as default.

v2.16

14.11.2022, v2.16:

Bugfixes:

• format mz: in the binary map of listings, the RVA of segments belonging
  to a group may have been incorrect, since the group's align adjustment
  was ignored; see grpalign.asm.
• symbol $ is internally marked as assembly-time variable - this caused
  fixup problems in OMF format; see fixup6.asm.
• invoke directive, 32-bit code, "ADDR": if the address of a symbol
  in a 16-bit segment was to be pushed, just a WORD was pushed ( Masm is
  buggy, pushes a DWORD, then a WORD ); see invoke41.asm.
  In the "reverse" case, that is, if the address of a symbol in a 32-bit
  segment was to be pushed as a WORD in a 16-bit code segment, a DWORD
  was pushed instead; see invoke43.asm.
• invoke directive, 32-bit code, "OFFSET": if the address of a symbol
  in a 16-bit segment was to be pushed, just a WORD was pushed ( Masm emits
  an error in this case - this is now done as well ); see invoke42.asm.
  The "reverse" case ( 16-bit code, address of a symbol in a 32-bit
  segment to be pushed ) must also be rejected; see invoke44.asm.
• invoke directive, "ADDR", "far ptr" modifier for vararg arguments: the
  modifier was ignored, so that no segment part was pushed; see
  invoke45.asm & invoke46.asm.
• invoke directive, ptr types: invoke might have emitted error "argument
  type mismatch" for arguments of type MT_PTR if their distance didn't
  match the one from the memory model; see invoke47.asm - invoke52.asm.
• externdef directive: if the symbol was already defined but neither
  internal nor external, no error was emitted; see extern16.asm.
• pointer to data did ignore default distance of memory model; see
  dsize.asm.
• option -pe: .data and .data? sections weren't merged; see pe11.asm.
• option -pe: if a section with initialized data (.reloc) followed a .data?
  section, space for the .data? section was reserved in the binary; see
  pe11.asm.
• if the difference of 2 labels ( at least one is a forward reference ) is
  used as operand for another op ( shift ), error 'constant is expected'
  may have occured; see forward9.asm.
• option -pe: binary without relocations got too long because the (empty)
  .reloc section wasn't removed; see pe12.asm.

Other changes:

• option -pe: allow imports by number; see pe10.asm.
• option -Zg: increased Masm compatible codegen for hll expressions; see
  hll4.asm.
• warning "Prologue not at PROC start" (level 3) added if the procedure's
  prologue isn't generated at the procedure's start address; see
  invoke53.asm.
• option -pe: default image bases changed to 0x10000000 for PE32 dlls,
  0x140000000 for PE64 exes, 0x180000000 for PE64 dlls (MS link compat).
• error 2058 ( "magnitude of offset exceeds 16-bit" ) changed to a warning
  ( level 3 ); see offset15.asm & offset16.asm.
• if option -Zg is on, OPATTR is to return language bits for procs/protos
  only; see opattr10.asm.

v2.15

Bugfixes:

• format bin, mz, omf, pe: fixup created by direct addressing without
  symbol ( _DATA:1234h ) was still not always correct if segment was in a
  group; see lea2.asm.
• invoke directive: a missing left square bracket for indirect addressing
  may have been accepted if the size of the register matched the expected
  size; see invoke4.asm.
• using long instead of intptr_t to hold result of _findfirst() caused a
  GPF in Win64 version; see main.c.
• erroneously emitted error 'Magnitude of offset exceeds 16-bit' in direct
  addressing mode without symbol if segment register had no assume, segment
  was 32-bit and default word size was 16-bit only; see offset14.asm.
• prologue argument was ignored if procedure had no parameters and
  locals; see loadds.asm.
• text macro and macro function expansion might have caused a buffer
  overflow if expanded line got too long; see expans42.asm & expans44.asm.
• macro function expansion inside text macros had swallowed trailing white
  spaces behind the macro invokation, sometimes resulting in errors; see
  expans43.asm.
• a constant defined via EQU could have been changed to an assembly-time
  variable if the value didn't change; see equate30.asm.
• if two relocatable items belonging to different segments were
  subtracted in an expression on the right side of EQU, strange error
  "symbol redefinition" was emitted when that symbol was used later; see
  equate31.asm.
• assembly-time variables: if the expression on the right side of the
  equal sign contained an undefined symbol, it might have been ignored;
  see expr7.asm.
• -pe option: the name in the export directory may have started with a
  slash/backslash.

Other changes:

• error "too many segments/sections defined..." emitted for omf and coff
  if the number of segments/sections cannot be handled by those formats.
• warning "64-bit label referenced with 32-bit fixup", introduced in v2.14,
  will be emitted at level 4 only if the code referencing the label isn't
  64-bit.
• format coff: warning "Output format doesn't support anonymous fixups..."
  displayed if a segment override other than FLAT: is used without symbol.
• .model flat directive in 64-bit is now rejected in omf format.
• warning displayed in binary formats if a linker directive section is
  detected.
• warning displayed for PUSHW/PUSHD if operand is a segment register and
  the size suffix doesn't match the current mode; see pushwsr.asm and
  pushdsr.asm.
• if user-defined prologue returns 0, offsets of local variables are no
  longer "adjusted". This is undocumented Masm behavior; see prolog3.asm.
• directive OPTION READONLY will no longer cause an error, just a warning
  that it's ignored.

v2.14

20.12.2020, v2.14:

Bugfixes:

• regression in v2.13: if a not yet defined segment was added to a group,
  it got the current offset size, which caused an error if this wasn't the
  offset size when the segment was created; see group5.asm.
• there was still the possibility that segments with different offset
  sizes were added to the same group; see group6.asm.
• string operation MOVSx with operands: if destination got an override
  (ES:), a source without override was rejected; see string4.asm.
• string operation MOVSx with operands: destination with override
  != ES wasn't rejected; see string5.asm.
• string operations with symbolic memory operands did ignore assumes, thus
  no segment prefix was created; see string8.asm.
• comment behind MACRO directive was appended to all lines inside the
  macro; see lst2.asm.
• jwasmr.exe: memory corruption may have caused a system crash if pass one
  ended with an error.
• some errors occured in LP64 systems (linux, unix) because LONG_MAX and
  LONG_MIN were used inadequately.
• INVLPG with direct address >= 0x10000 in 16-bit code was rejected, even
  if preceded by 'assume ds:flat'; see invlpg2.asm.
• using direct addresses without symbol ( i.e. flat:[1234h] ) may have
  caused a GPF if used in math opcodes ( CMP, OR, ...); see flatgrp2.asm.
• option -bin: using direct addressing without symbol ( _DATA:1234h )
  didn't store the offset in the binary; see data1.asm.

Other changes:

• jfc.asm translated to jfc.c so it can be used in both 32- and 64-bit
  unix systems.
• added runtest.sh, clrdir.sh and compress.sh to allow regression test in
  unix systems. A few tests are skipped, since jwasmr.exe and MS link.exe
  won't run in those systems.
• DJGPP: djgpp apparently doesn't know "_stat", needs "stat" instead; see
  omf.c
• error "ADDR32 fixup to at location ..." emitted if 64-bit PE is
  created and 32-bit fixup target address is >= 0x80000000; see pe7.asm.
• warning "64-bit label referenced with 32-bit fixup" ( warning level 3 )
  emitted if 32-bit address used for 64-bit target.
• translation for 8/16/32-bit register arguments for INVOKE in 64-bit if
  they are to be pushed; see invok64d.asm.

v2.13

Bugfixes:

• regression in v2.12: option -bin didn't work if binary started with
  ORG != 0 ( i.e. DOS COM file format ); see bintest2.asm.
• option -mz,-bin: invalid binary was written if a .bss or .stack segment
  was followed by a segment with initialized data; see mz2.asm.
• options -mz,-bin,-omf: a segment override may have caused a group
  association to be ignored; see fixup5.asm.
• options -mz,-bin: simplified segment directives now generate an assume
  for CS that refers to the group (if the segment is in a group).
  This solves some fixup problems; see fixup5a.asm.
• INVOKE directive, 64-bit: if the calling convention wasn't win64
  (=fastcall), an address argument was translated to a "push offset ",
  which didn't work in 64-bit code; see invok64c.asm.
• option -pe: if a DOS stub was added in the source (section .hdr$1), it
  got the wrong segtype, resulting in an invalid binary; see pe4.asm.
• stack offsets may have caused an error in 64-bit mode or a warning in
  the other modes if an overflow occured; see stkdisp1.asm.
• option -pe,-bin: relative fixups weren't handled correctly in all cases;
  was relevant for 64-bit only; see pe5.asm.
• if an 'undefined symbol' error was displayed, a wrong source file was
  given if the error occured in a macro and the line was the first 'real'
  code line; see undef1.asm.
• a qword-sized parameter in an invoke wasn't pushed correctly if a .x64
  directive was in the source, although the current mode was not 64-bit;
  see invoke40.asm.
• equates in listing weren't always displayed correctly if value exceeded
  32-bit; see lst1.asm.
• ensured that if cpu is 64-bit, .model is restricted to flat.
• TYPE operator didn't work with register operands inside square brackets;
  see types16.asm & types17.asm.
• in 64-bit mode, 16-bit index register wasn't rejected; see expr6.asm.
• assembly-time variables modified by directives SIZESTR/INSTR did set
  the lower 32-bits of the value only; see equate29.asm.
• option -pe: when arranging segments to sections, the segment names
  defined in the source were used; this may have given unexpected results
  if segment names were translated to section names ("_TEXT" to ".text" );
  see pe6.asm.
• hll directives: an internally created label was "swallowed" in some cases
  if operators && and || were used in combination with brackets; see
  rtcond10.asm - rtcond16.asm.
• check for segment word size change wasn't as strict as Masm's. If a not
  yet defined segment was added to a group, no consistency check was done
  later when the segment was actually defined; see group4.asm.
• 32-bit relative fixups were created only when the target was 64-bit. Now
  they were created generally when the current segment is 64-bit. This
  makes a difference for mixed-mode binaries ( -mz and -bin format ),
  because it simplifies access of variables in 16-/32-bit segments from
  64-bit; see relfix64.asm.

Other changes:

• project tree changed so all generated files are located in 'build'
  subtree, and source files are in 'src'.
• -Sz cmdline option added to suppress assembly result in listing. This
  was mostly done to simplify comparing listing files in regression tests.