Notes about MS-DOS executables and compilers:
- Borland start-up code is reported to switch the screen mode auto-
matically if it's not 80 columns (or possibly 40) and either 25, 43
or 50 lines. In particular, extended modes such as 100x40 are not
retained.
- Borland start-up code also uses interrupt 1Ah, causing incorrect
behavior (including lock-ups) on some Japanese MS-DOS machines such
as the Fujitsu FMR series, which lack this interrupt.
- Some(?) Borland compilers are apparently incapable of putting static
data into far memory; this means all of UnZip's strings are in near
memory, and there is not enough room to enable ZipInfo in the small
memory model. The medium memory model is the default for now, but
it may be necessary in some cases to use the large model.
- Older Borland compilers do not understand source files with Unix
line-endings (LF rather than CR/LF). Use "flip" or a similar utility
to convert the line endings before compiling.
- The Borland 5.00 compiler is simply too buggy to use on WizUnZip, both
16-bit and 32-bit versions, and we recommend avoiding it for now even
on the commmand-line version of UnZip.
- Microsoft C 5.1 large-model code is more than an order of magnitude
slower than the identical code compiled with MSC 6 or 7 (a factor of
15 in our tests, actually). This may be due to a lousy optimizer or
lousy libraries; regardless, since UnZip is hovering at the doorstep
of the large memory model, we recommend upgrading to a later version
of the compiler.
For these reasons, Info-ZIP's distributed versions of the 16-bit MS-DOS
executables are compiled with MSC 6 or 7.
- djgpp 2.x (currently 2.03) is no longer distributed with the go32 extender.
Instead, a 2K stub bound into the executable searches for a DPMI server;
if none is found, it loads the default DPMI server while executing UnZip.
Both djgpp 1.x and 2.x are capable of substituting a list of files in an
ASCII file (say, `foo') on the command line; for example, "unzip archive
@foo" will extract from `archive' all of the files listed in `foo'. Note,
however, that djgpp 2.x is considerably slower than 1.x at file extraction
(roughly twice as slow, in fact); see proginfo/perform.dos in the UnZip 5.3
source distribution for details.
- djgpp 2.0's long-filename support is somewhat flaky; users should upgrade
to version 2.01 instead.
- The default wildcard ("globbing") behavior of djgpp 1.x/go32 is disabled
by default in UnZip, but this can be overridden if the GO32 environment
variable is set to "glob". This will cause UnZip to fail with various
odd errors about "filename not matched" and the like; to avoid this, set
the GO32 variable to "noglob" or unset it altogether. (The documented
method of avoiding this by quoting wildcards with single quotes was
buggy in djgpp 1.11 but is reported fixed in 1.12; not tested.)
- djgpp 1.x's handling of timezones, necessary for the correct conversion of
MS-DOS file times to those used in the Unix-like C library, is completely
broken in djgpp 1.12 and probably earlier versions as well. It is fixed
(or very close to it) in the 1.12m4 patch release and reportedly in the
2.x series, so be sure to use one of those when compiling. Otherwise
UnZip's -f and -u (freshen/update) functions will not work correctly.
It is reportedly necessary to set the TZDIR environment variable correctly
with 1.12m4; for example, add `set TZDIR=c:/djgpp/zoneinfo' or similar to
autoexec.bat.
- djgpp 1.x/go32 executables, when run in a DOS box under OS/2 *and* extrac-
ting to an HPFS disk *and* overwriting existing files (intentionally, that
is), do not set the files' timestamps correctly. Instead, the timestamps
remain set to whatever the original files' stamps were. This is a pretty
obscure bug, but it does *not* occur in the 16-bit version so it seems
to be go32's fault somehow.
- According to notes found in another package, there was a known conflict
between djgpp 1.x's go32 extender and QEMM's DPMI; this was apparently
fixed in QEMM 7.04/QDPMI 1.05, but if you still have an older version
(1.03 or 1.01), add "set GO32=nodpmi" to your autoexec.bat to avoid the
conflict.)
- [For Zip only, the djgpp/go32 extender goes nuts with the copying and/or
deletion of some sort of a temporary file (swap file?) after compression
is finished; this can take 30 seconds or more and really hurts perfor-
mance. It doesn't affect UnZip, apparently.]
- [Also apparently for Zip only, djgpp/go32 is reported to have problems
when EMM386 is set to NOEMS; it sometimes gives the error message, "CPU
must be in REAL mode (not V86 mode) to run this program without VCPI.
(If you are using an EMS emulator, make sure that EMS isn't disabled)"
Sometimes Zip works correctly, however, possibly due to other software
having been run previously.]
- emx+gcc's DOS extender does not understand DPMI, and while there is an
alternative extender called RSX available (found in dpmigcc4.zip as of
August 1994), its setup is somewhat kludgy when the local memory manager
supports both DPMI and VCPI (or something else). It's also not yet as
widely known or available as djgpp.
- The free PMODE/W extender, used in conjunction with executables compiled
with Watcom C 10.x and run in an OS/2 DOS box, appears to use up some
critical DPMI resource and will fail to run after a few dozen executions
("PMODE/W: DPMI error" and/or SYS 3176). Some newer versions of PMODE/W,
in combination with "unzip -v" on certain zipfiles (e.g., APMTST.ZIP from
IBM/EWS), fail immediately (SYS 3176). And on some OS/2 systems, *any*
use of the PMODE/W executables causes the machine to lock up.
- PMODE/W is also reported to lock up pure DOS systems if QEMM is running.
- At least older versions of PMODE/W, used in conjunction with Microsoft's
EMM386, cause UnZip to start up extremely slowly. (This problem does not
occur with QEMM.)
For these reasons Info-ZIP's distributed 32-bit MS-DOS executables will
be compiled with djgpp 2.03, mainly because of its nice long-filename
support when running in a Win32 DOS box. The cwsdpmi DPMI server will be
bundled if necessary.
SPC 2000-04-16
