This is a list of all official ports of third party software to Sortix and their current status. Each port has meta data and a list of issues. Issues in the past tense have been patched around, otherwise they are remaining issues. Ports are of course always behind the official upstream versions and issues may have been fixed upstream
Our philosophy is that software should strive to work well on unknown operating systems. Sortix does not upstream patches that specifically mention/support Sortix. Sortix is perfectly happy maintaining its own patches for specific Sortix support when it must. However you are more than welcome to help upstream patches that make software more portable in general, but remember not burden upstream projects with specific Sortix workarounds.
Each port has the following meta data:
Field
Meaning
Status
An assessment of code quality and issues porting it to Sortix:
Complicated: See notes. Alright: Considerable issues. Good: Minor issues. Great: Exemplary port. Sortix: Maintained by Sortix. OpenBSD: Imported from OpenBSD. Wrapper: Wrapper for another source package.
Native
Whether it builds natively on Sortix.
Updated
Date of last update to the port.
Version
The version of the software.
Released
When the software version was released.
Upstream Version
(If a Sortix fork) The upstream version forked from.
Upstream Released
(If a Sortix fork) When the upstream release was made.
This is really Sortix binutils, a local fork of GNU binutils enhanced to know about Sortix.
State in the binutils, gcc and libstdc++ port patches that this is not the real patches; that this is based on sortix project foo rather than upstream foo.
Rework how the tooldir problem is solved in the binutils patches, or at least consider the problem for a while.
Uses the obsolete clock() function internally and libiberty simulates it.
Uses utime and <utime.h> and isn't able to use utimens, though that has been patched in for Sortix support.
binutils/gas/config/tc-tic54x.c uses the unsafe tmpnam.
libiberty provides its own tmpnam.
Needs to depend on libintl rather than gettext.
Native build has trouble with documentation and lack of texinfo due to timestamps being not being made correctly when making sortix-binutils. Reseting all file timestamps to the same time fixes the problem.
Intrusively uses gnulib and has no faith in the standard library. It runs a lot of paranoid tests that might succeed and the standard library gets used if lucky. If unlucky or cross-compiling, it doesn't assume the best about the operating system, but the worst and all sorts of nasty fallback code gets compiled in. (This is somewhat patched around using nasty gnulib configure cache variables)
Doesn't make distclean properly.
Cross-compiles programs and tries to run them.
Ships lib/xmalloc.c that provides xcalloc, but the gl_FUNC_CALLOC_GNU m4 test that defines HAVE_CALLOC_GNU is not part of the configure script, so it doesn't listen for our cache variable; thus the calloc implementation isn't trusted to do overflow checks, xcalloc does its own testing as well. (This port needs to be updated to work around this.)
Ships lib/timevar.c (of gcc descend) that uses functions like getrusage, timens, and clock. This needs an audit and potentially be upgraded to the superior Sortix time reporting functions.
Ships lib/xmalloc.c that provides xcalloc, but the gl_FUNC_CALLOC_GNU m4 test that defines HAVE_CALLOC_GNU is not part of the configure script, so it doesn't listen for our cache variable; thus the calloc implementation isn't trusted to do overflow checks, xcalloc does its own testing as well.
This is really Sortix gcc, a local fork of GNU gcc enhanced to know about Sortix.
State in the binutils, gcc and libstdc++ port patches that this is not the real patches; that this is based on sortix project foo rather than upstream foo.
Bundles with its own libz.It is used by default unless the user explicitly says no.
The gcc/libsanitizer/tsan/tsan_symbolize_addr2line_linux.cc and gcc/libsanitizer/sanitizer_common/sanitizer_symbolizer_posix_libcdep.cc have scary getdtablesize calls. Actually those files are pretty scary. This looks like a place security issues could pile up.
Uses the obsolete caddr_t type.
Potential conflict with TIME_MAX in ports/gcc/libitm/method-ml.cc.
Ships timevar.c that uses functions like getrusage, timens, and clock. This needs an audit and potentially be upgraded to the superior Sortix time reporting functions. (Mind also this has been copied into gnulib)
Uses the obsolete clock() function internally and libiberty simulates it.
Bundles its own libffi.
Uses utime rather than utimens in libjava.
Uses utime rather than utimens in ada code.
Might be using the dangerous -funsafe-math-optimizations in libgo.
_eprintf needs to be deleted from libgcc. Or, perhaps, as a better compromise: Check whether fprintf, and fflush exists before using them.
Remove fsf-funding.7.
Add to tixbuildinfo that pkg.configure.with-sysroot-ld-bug=true so tix-build don't have to default to this.
Problems with libintl and libiconv linking order. (Patched)
Deletes the configure script on distclean even though it's included in releases. (Patched)
The configure script doesn't cross-compile as two checks needs to be run: A check for whether snprintf has a bogus return value, and whether you can fread directories. It should instead assume the best for unknown platforms when cross-compiling, config.mak.uname can be used to blacklist broken systems. (Patched)
HAVE_DEV_TTY is only defined config.mak.uname based on the build system and is not set by configure. (Patched)
Perl stuff is installed in /usr/local and not exactly sure where that path comes from. (Patched).
Progress bars requires restartable system calls which Sortix doesn't have right now. (Patched) (__SORTIX_HAS_RESTARTABLE_SYSCALLS__)
Replacement hstrerror is compiled (and doesn't compile) even if it is not used. (Patched).
TIME_MAX constant conflicts with Sortix constant. (Patched).
Single use of rand replaced with arc4random to avoid compile warning. (Patched).
Manual pages were not build & installed by default
Pthread cancellation is used. (Patched).
Uses variables like SHELL_PATH, PERL_PATH, PYTHON_PATH at both build-time and run-time. However, they need to be distinct in the case of cross-compiling. Perhaps switch this to a SHELL_PATH_FOR_BUILD or SHELL_PATH_FOR_HOST. (Not patched)
Why does configure.ac GIT_CHECK_FUNC run AC_SEARCH_LIBS when it finds the function, I would expect it to do that when it fails to find it. Is it because of race conditions? Is it a nop, because it searches no libraries? (Not patched)
Used curl-config instead of pkg-config to detect if curl supported SSL. (Patched)
Has a single stray raw pkg-config call in configure.ac that should have been with PKG_CONFIG instead.
Likes to use link-time optimization automatically if it naively appears to work, even though it doesn't really and makes my Sortix toolchain explode (though, that's my fault).
Really doesn't believe X is unavailable unless a million configure options are passed.
configure required gethostbyname to exist even if it's not used.
configure "amended" the distclean targets, but it corrupted the Makefile with hundreds of thousands of duplicate rm calls.
struct hostent needed to be declared even though the gethostbyname family is not used.
Used sprintf instead of snprintf.
The tarball wasn't distclean.
The distclean target didn't fully distclean.
configure needs to be told --with-ca-bundle=/etc/ssl/cert.pem --with-ca-path=/etc/ssl/certs as it otherwise can't locate the libssl certificate storage when cross-compiling.
freetype2.pc has a HTTP URL instead of HTTPS. (Patched)
Ships a cross-unsafe freetype-config program. (Patched)
Uses sprintf instead of snprintf. (Patched)
Contains bundled copy of zlib that is used when the system has no libz. (Patched)
valid_entries triggers maybe initialized warning. (Patched to initialize with 0)
Can't decide where its headers should be installed, so provides a <ft2build.h> header that #define's the real locations.
Including the real headers doesn't work because they check if the user actually used the <ft2build.h> header, then it includes the header again for good measure.
Then wraps its headers in another freetype2/ directory and forces the user to locate it with pkg-config.
Doesn't trust the standard library and wraps everything with FT_ prefixes so it becomes effectively written in the FT_C programming language.
Probably lots of other situations when examined closer.
Doesn't properly check is fpclassify works, I tried to upstream a patch that fixes that, don't think it made it into the official version, need to try again.
(Likes to regenerate makefiles a bit too much, or is that my fault?)
Assumes the worst when in a few cases when cross-compiling and unable to test. For instance that NAN != NAN. I get a vibe that it does work work safely around this, but it's worth investigating.
Port is set to do out-of-directory builds but that isn't needed. Remove this during the next package update.
Uses the obsolete clock() function in the examples.
Native build requires makeinfo for documentation.
Native build has a bug in the makefile's last line, it includes *.d files but there's no leading dash so it can fail. Can be fixed with --disable-dependency-tracking or fixing configure DEP_INCLUDE to -include.
Some weird issues with <ogg/config_types.h> being generated and including the <stdint.h> headers (and others) with the name substituted in at configure time.
This is really Sortix libstdc++, a local fork of GNU libstdc++ enhanced to know about Sortix.
State in the binutils, gcc and libstdc++ port patches that this is not the real patches; that this is based on sortix project foo rather than upstream foo.
There are probably many exciting things to discover if you study the code.
Uses the obsolete clock() function in testsuite/util/performance/time/elapsed_timer.cc.
Doesn't detect that thread local storage is available on Sortix.
GLIBCXX_ENABLE_LIBSTDCXX_TIME isn't friendly for unknown operating systems, and this also isn't friendly for musl.
runtest and testrecurse have problems with sortix not having glob yet.
Tries to use python2.7-config scripts instead of pkg-config.
Uses the obsolete clock() function in a few files.
What's this with xmlChar?
The port spews a bunch of compiler warnings.
Why nanohttp and nanoftp exist and why are they even built? Is this a ftp and http client sublibrary? Is this thing full of security issues because it's ill-known code?
It uses ansidecls.h if available in its public headers.
Is able to link with liblzma but doesn't put that in the pkgconfig file.
Required knowledge about the host operating system.
Only cross-compiled to white-listed systems.
configure may detect libiconv needed to be linked with in the wrong order.
Assumed getaddrinfo is buggy when cross-compiling.
Some problems if the host operating system doesn't do dynamic linking.
Funky semantics regarding what version suffixes the installed python programs get.
Removed python-config.
Looks like it might use POSIX advisory locks.
Had trouble with the empty prefix being the root directory at runtime?
Had a use ofputenv not wrapped in HAVE_PUTENV.
Used the obsolete gethostbynameand gethostbyaddrinterfaces.
Checked for /dev/ptmx and /dev/ptxc during configure (and failed to cross-build) even though posix_openpt exists, and openpty() ends up being used anyways.
Bundles its own libz.
python/Modules/socketmodule.c hard-codes SOL_TCP to 6 and SOL_UDP to 17 instead of using the standard IPPROTO_TCP and IPPROTO_UDP constants. Some Python test code also uses the SOL_TCP constant.
Uses the obsolete clock() function in a few files.
Bundles its own libffi.
Pip stuff and such gets installed into /usr and might be downloaded. Investigate ensurepip. Disabled this for now.
LDSHARED and LDCXXSHARED defaulted to ld (rather than $CC -shared) on unknown operating systems.
Patched to use arc4random rather than urandom.
Modules/Setup.dist changed to static link as many modules as possible
ssl module not static linked as static modules can't be optional.
libcurses module not static linked as static modules can't be optional.
Bootstrap phase for cross-builds as the exact same python version must be installed in order to cross-compile.
Assumed MAXPATHLEN existed for the purpose of realpath.
Native build encounters weird problems in importlib.
