64-bit transition part 12: ksh93 lexing, parsing and subshells

[JohnoKing]

This is the twelfth of the thickfold patch series, which enables ksh93 to operate within a 64-bit address space.

The parts of ksh93 affected by this commit are:

• The lexing and parsing components in lex.c, parse.c, fcin.c and trestore.c.
  • Added an assert() after stkalloc() to fix the following LTO warning:

warning: 'memset' writing 60 bytes into a region of size 0 overflows the destination [-Wstringop-overflow=]
      1819 |                         memset(ioq,0,sizeof(*ioq));
           |                         ^
    lto1: note: destination object is likely at address zero

• Minor fixes for the virtual subshell mechanism.
• The code underlying shcomp, aka sh_tdump().
• A minor fix to a cast in sh_timeradd().

Change in the number of warnings on Linux when compiling with clang using -Wsign-compare -Wshorten-64-to-32 -Wsign-conversion -Wimplicit-int-conversion: 481 => 352 => 37 (progression from part 11 => part 12 => part 13)

Progresses #592

[McDutchie]

Added an assert() after stkalloc() to fix the following LTO warning:

Sorry but I don't want to add asserts to silence incorrect compiler warnings, in part because asserts should be getting compiled out on release builds, in part because that's not what asserts are for.

I get why you've been adding a lot of typecasts to silence warnings in extra-paranoid warning modes – it helps you isolate the warnings that tell you where variable types need to change to accommodate large objects in 64-bit address spaces. But this particular warning here is simply wrong, and this isn't the right way to silence it.

If it really annoys you, fair enough; you could instead add a #pragma to disable the warning, with a short comment explaining the reason for adding it.