The Unix and Windows linkers appear to simply ignore if any symbol is
defined multiple times in different object files and libraries.
The VMS linker, on the other hand, warns about it, loud and clear. It
will still create the executable, but does so screaming. So we
complicate things by saving the linker output, look through all the
errors and warnings, and if they are only made up of %LINK-W-MULDEF,
we let it pass, otherwise we output the linker output and raise the
same exit code we got from the linker.
Reviewed-by: Emilia Käsper <emilia@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1789)
Instead of deliberately leaking a reference to ourselves, use nodelete
which does this more neatly. Only for Linux at the moment.
Reviewed-by: Tim Hudson <tjh@openssl.org>
VC-noCE-common and VC-WIN64-common were missing this line:
template => 1,
Fixes GH#1809
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1811)
Move manpages to manX directories
Add Windows/VMS install fix from Richard Levitte
Update README
Fix typo's
Remove some duplicates
Reviewed-by: Richard Levitte <levitte@openssl.org>
The current version of the VMS compiler provides C99 features,
strictly language wise. Unfortunately, even the most recent standard
library isn't fully updated for that standard, so we need to use an
earlier standard that the compiler supports.
Most importantly, this affects the __STDC_VERSION__ value, which the
compiler unfortunately currently defaults to 199901L. With this
change we won't have to give VMS special treatment when looking for
features based on that macro.
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1785)
Make Configure recognise -rpath and -R to support user added rpaths
for OSF1 and Solaris. For convenience, add a variable LIBRPATH in the
Unix Makefile, which the users can use as follows:
./config [options] -Wl,-rpath,\$(LIBRPATH)
Reviewed-by: Rich Salz <rsalz@openssl.org>
Before OpenSSL 1.1.0, binaries were installed in a non-standard
location by default, and runpath directories were therefore added in
those binaries, to make sure the executables would be able to find the
shared libraries they were linked with.
With OpenSSL 1.1.0 and on, binaries are installed in standard
directories by default, and the addition of runpath directories is
therefore not needed any more.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Note that it relies on a trick from Configure, where file names for
object files made from C++ source get '.cc' replaced with '_cc.o' to
recognise them. This is needed so the correct compiler is used when
linking binaries.
Reviewed-by: Rich Salz <rsalz@openssl.org>
A note: this will form object file names by changing '.cc' to
'_cc.o'. This will permit other configuration code to recognise these
object files were built for C++ rather than C.
Reviewed-by: Rich Salz <rsalz@openssl.org>
This is an amendment to the september 8 commit titled "VMS: Don't
force symbol mixed case when building DSOs"
Reviewed-by: Rich Salz <rsalz@openssl.org>
This flag got moved after -xarch=v9 in 1.1.0 and had the unexpected
side effect of the compiler building for 32-bit v8plusa instead of v9.
GH#1521
CLA: none; trivial
Signed-off-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
This is generalised by having the following macros for stuff that won't
be installed:
NO_INST_LIB_CFLAGS, used instead of LIB_CFLAGS
NO_INST_DSO_CFLAGS, used instead of DSO_CFLAGS
NO_INST_BIN_CFLAGS, used instead of BIN_CFLAGS
They take values from corresponding target config fields if those are
defined, otherwise they take the respective values from LIB_CFLAGS,
DSO_CFLAGS and BIN_CFLAGS.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Send a bit information to the build file template functions. For
src2obj(), the additional option 'product' holds the name of the final
file that the object file will go into. Additionally, the diverse
functions will get the option 'installed', with a value that evaluates
true if the final product is to be installed, otherwise false.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Improve interchangeability of aix*-gcc targets by linking shared
libraries with -static-libgcc, and address linking problems with
vendor compiler.
Reviewed-by: Richard Levitte <levitte@openssl.org>
The background story is that util/shlib_wrap.sh was setting LD_PRELOAD
or similar platform dependent variables, just in case the shared
libraries were built with -rpath. Unfortunately, this doesn't work
too well with asan, msan or ubsan.
So, the solution is to forbid the combination of shared libraries,
-rpath and any of the sanity analyzers we can configure.
This changes util/shlib_wrap.sh so it only contains the code that sets
LD_PRELOAD when -rpath has been used when configuring.
Reviewed-by: Rich Salz <rsalz@openssl.org>
In the case of using an independent makedepend, we had split that into
two separate recipes, one depending on the other. However, there are
cases where the makedepend recipe was always trying, but doesn't
update the time stamp of the .d file because there are no actual
changes, and thereby causing constant updates of the object files.
This change makes one recipe that takes care of both makedepend och
cc, thereby avoiding these extra updates.
Reviewed-by: Andy Polyakov <appro@openssl.org>
MIPS[32|64]R6 is binary and source incompatible with previous MIPS ISA
specifications. Fortunately it's still possible to resolve differences
in source code with standard pre-processor and switching to trap-free
version of addition and subtraction instructions.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Since vendor assembler can't assemble our modules with -KPIC flag,
it, assembly support, was not available as an option. But this
means lack of side-channel resistant code, which is incompatible
with security by todays standards.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Because some targets execute perl code that might die, we risk
incomplete lists. Make it so dying doesn't happen when we're listing
targets.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Most of the time, this isn't strictly needed. However, in the default
extern model (called relaxed refdef), symbols are treated as weak
common objects unless they are initialised. The librarian doesn't
include weak symbols in the (static) libraries, which renders them
invisible when linking a program with said those libraries, which is a
problem at times.
Using the strict refdef model is much more like standard C on all
other platforms, and thereby avoid the issues that come with the
relaxed refdef model.
Reviewed-by: Rich Salz <rsalz@openssl.org>
The options RC4_CHUNK_LL, DES_PTR, and BF_PTR were removed by Rich
in commit 3e9e810f2e but were still
sticking around in a coupule configuration entries.
Since they're unused, remove them.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1390)
Experience shows that pod2html changes directory during its process
without properly adjusting the given source directory.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Instead, install the new one as openssl.cnf.dist (openssl.cnf-dist on
VMS), and only install it as openssl.cnf if that file doesn't already
exist.
Also, don't install with exec privileges on VMS.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Windows never composes UTF-8 strings as result of user interaction
such as passing command-line argument. The only way to compose one
is programmatic conversion from WCHAR string, which in turn can be
picked up on command line.
[For reference, why not wmain, it's not an option on MinGW.]
Reviewed-by: Richard Levitte <levitte@openssl.org>
The way it was implemented before this change, the shared libraries
were installed twice. On a file system that supports file
generations, that's a waste. Slightly rearranging the install targets
solves the problem.
Reviewed-by: Rich Salz <rsalz@openssl.org>
On non-Windows platforms, shared libraries are both development and
runtime files. We only installed them as development files, this
makes sure they get installed as runtime files as well.
Reviewed-by: Rich Salz <rsalz@openssl.org>
This adds a new target 'build_programs' and makes 'build_apps' and
'build_tests' aliases for it, for backward compatibility.
Reviewed-by: Rich Salz <rsalz@openssl.org>
With a number of tools, especially those coming with Visual Studio,
some command options are separated from their argument with a space,
others with a space. Since we parametrise them, we can't know
beforehand which it will be, so we must allow the input and output
options to have either.
However, spaces at the end of nmake macro values are trimmed, so allow
spaces to exist by adding a reference to an undefined macro at the end.
Reviewed-by: Andy Polyakov <appro@openssl.org>
This is only done for the platforms where 'OPENSSL_USE_APPLINK' is defined.
Also, change the docs of OPENSSL_Applink to say where to find applink.c
in the installation directory.
Reviewed-by: Rich Salz <rsalz@openssl.org>
The easiest way to take care of manifest files is to integrate them
into the associated binary (.exe or .dll). MT (the Manifest Tool) is
the utility to use for this.
Reviewed-by: Rich Salz <rsalz@openssl.org>
With OpenSSL 1.1 and on, the engines are tightly tied to the shared
library they're to be used with. That makes them depend on the
pointer size as well as the shared library version, and this gets
reflected in the name of the directory they're installed in.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Some Unix variants require shared libraries to have the execute
permissions set, or they won't be loadable or executable when loaded.
Among others, cygwin has this requirement.
Reviewed-by: Matt Caswell <matt@openssl.org>
$openssldir and $enginesdir were mistakenly made unavailable to other
perl fragments. They are still needed in the definition of CFLAGS.
Reviewed-by: Rich Salz <rsalz@openssl.org>
OpenSSL engines are tied to the OpenSSL shared library versions,
starting with OpenSSL 1.1. We therefore need to install them in
directories which have the shared library version in it's name, to
easily allow multiple OpenSSL versions to be installed at the same
time.
For VMS, the change is a bit more involved, primarly because the top
installation directory was already versioned, *as well as* some of the
files inside. That's a bit too much. Version numbering in files is
also a bit different on VMS. The engines for shared library version
1.1 will therefore end up in OSSL$INSTROOT:[ENGINES0101.'arch']
('arch' is the architecture we build for)
Reviewed-by: Rich Salz <rsalz@openssl.org>
OpenSSL engines are tied to the OpenSSL shared library versions,
starting with OpenSSL 1.1. We therefore need to install them in
directories which have the shared library version in it's name, to
easily allow multiple OpenSSL versions to be installed at the same
time.
For Unix, the default installation directory is changed from
$PREFIX/lib/engines to $PREFIX/lib/engines-${major}_${minor} (mingw)
or $PREFIX/lib/engines-${major}.${minor} (all but mingw)
($PREFIX is the directory given for the configuration option --prefix,
and ${major} and ${minor} are the major and minor shared library
version numbers)
Reviewed-by: Rich Salz <rsalz@openssl.org>
OpenSSL engines are tied to the OpenSSL shared library versions,
starting with OpenSSL 1.1. We therefore need to install them in
directories which have the shared library version in it's name, to
easily allow multiple OpenSSL versions to be installed at the same
time.
For windows, the default installation directory is changed from
$PREFIX/lib/engines to $PREFIX/lib/engines-${major}_${minor}
($PREFIX is the directory given for the configuration option --prefix,
and ${major} and ${minor} are the major and minor shared library
version numbers)
Reviewed-by: Rich Salz <rsalz@openssl.org>
When creating the library $lib.olb, make sure the extension is there.
Otherwise, a logical name with the same name as the file in question
will redirect the creation elsewhere.
Reviewed-by: Tim Hudson <tjh@openssl.org>
On VMS, it's customary to have a procedure to check that the software
was installed correctly and can run as advertised.
The procedure added here is fairly simple, it checks that all
libraries are in place, that the header crypto.h is in place, and that
the command 'openssl version -a' runs without trouble.
Reviewed-by: Rich Salz <rsalz@openssl.org>
- The install top is versioned by default. However, only the major
version should be used.
- the default areas for certs, private keys an config files have
changed, now all prefixed with 'OSSL$'. This gets reflected in
cryptlib.h.
- [.VMS]openssl_startup.com.in had some faults regarding creating
rooted concealed logical names.
Reviewed-by: Rich Salz <rsalz@openssl.org>
This makes it possible for script writers to lock on to a specific
version if they need to. Note that only the major version number is
used.
Reviewed-by: Rich Salz <rsalz@openssl.org>
$prefix was removed as part of the DESTDIR work. However, it was
still used to create the ENGINESDIR_dev and ENGINESDIR_dir variables,
so a restoration is needed.
Reviewed-by: Rich Salz <rsalz@openssl.org>
DESTDIR can't be used on Windows the same way as on Unix, the device
part of the installation paths get in the way.
To remedy this, have INSTALLTOP, OPENSSLDIR and ENGINESDIR get
different values depending on if $(DESTDIR) is empty or not, and use
$(INSTALLTOP), $(OPENSSLDIR) and $(ENGINESDIR) alone.
Reviewed-by: Andy Polyakov <appro@openssl.org>
This is just in case someone passed an inclusion path with the
configuration, and there are OpenSSL headers from another version
in there.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Since corresponding rule was removed from windows-makefile.tmpl out
of necessity, question popped if it's appropriate to harmonize even
unix-Makefile.tmpl. Note that as long as you work on single directory
'make lib<rary>.a' is effectively equivalent to 'make <dir/ectory>'
prior this modification.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Problem with Microsoft lib.exe is that it doesn't *update* modules
in .lib archive, but creates new one upon every invocation. As result
if a source file was updated and nmake was executed, a useless archive
with only one module was created. In other words one has to always
pass all .obj modules on command line, not only recently recompiled.
[This also creates dilemma for directory targets, e.g. crypto\aes,
that were added to simplify every-day life for developer. Since
whole idea behind those targets is to minimize the re-compile time
upon single file modification, the only sensible thing to do is to
omit intended library update.]
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
- User targets are now the same and generally do the same things
- configdata.pm depends on exactly the same files on all platforms
- VMS production of shared libraries is simplified
- VMS automatic dependency files get the extension .D rather than .MMS
Reviewed-by: Rich Salz <rsalz@openssl.org>
'DEPEND[]=file.h' becomes a special way to say that 'file.h' must be
generated before anything else is built. It's likely that a number
of source files depend on these header files, this provides a simple
way to make sure they are always generated even it the dependency data
hasn't been added to the build file yet.
Reviewed-by: Rich Salz <rsalz@openssl.org>
The directories for the final products were never registered, it was
plain luck that intermediary files were in the same place and
registered the directory anyway.
Also, scripts are generated directly from source (binaries go through
intermadiary object files), so we need to explicitely make sure to
avoid registering the source directory unless it's an in source
build.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
Previous build scheme allowed building just the stuff in one
subdirectory, like this:
make -C crypto/aes
Because the unified only has a top-level Makefile, this is not
possible with it. This change adds a replacement where each directory
we have something to build in becomes a target in its own right,
allowing building something like this:
make crypto/aes
The exception is the directory test, because we already have such a
target.
Reviewed-by: Stephen Henson <steve@openssl.org>
This applies when building out-of-source.
RT#4486
NOTE: we can't do the same for Unix, as Unix make doesn't handle this
type of issue. Also, directory specs are much less likely to have
spaces on Unix...
Reviewed-by: Matt Caswell <matt@openssl.org>
Someone wants to configure like this:
PERL="/usr/bin/env perl" ./config
The end goal is to get that in the #! line of CA.pl and a few other
scripts. That works well already, but in the Makefile, there were a
few lines looking like this:
PERL=$(PERL) $(PERL) whatever.pl ...
Those need some quoting.
RT#4311
Reviewed-by: Matt Caswell <matt@openssl.org>
Add Configure generated header files to $unified_info{generate}. This
makes sure the build files will pick them up with the rest for the
GENERATED macro, and thereby make sure they get cleaned away by 'make
clean'
Reviewed-by: Rich Salz <rsalz@openssl.org>
Some setups use links inside .git directory and make clean should not
remove them to avoid breaking git meta-information.
Signed-off-by: Cristian Stoica <cristian.stoica@nxp.com>
CLA: none; trivial
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1078)
DJGPP is a 3rd party configuration, we rely entirely on the OpenSSL to
help us fine tune and test. Therefore, it's moved to its own config.
Reviewed-by: Andy Polyakov <appro@openssl.org>
- some Perl versions are allergic to missing ';';
- don't stop if del fails;
- omit unused environment variable;
Reviewed-by: Stephen Henson <steve@openssl.org>
With Unixly Makefiles as well as with nmake, make variables are
transferred to the shell running the commands as envinronment
variables. This principle doesn't apply with MMS, so we must
explicitely define VERBOSE as commands when it's needed.
Reviewed-by: Rich Salz <rsalz@openssl.org>
If environment variables are not explanded early enough, expanded
strings are passed with single backslash to C compiler, e.g.
C:\Program Files, which effectively results in OpenSSL looking for
engines and certificates in C:Program Files.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Some non-Gnu compilers interpret -E -P combination differently.
some prioritize -E over -P, others -P over -E (in which case .i
file is generated and sometimes truncated because of redirection).
Reviewed-by: Richard Levitte <levitte@openssl.org>
When compiling all other C files, rely on the compiler to
automatically pick up the name translation information from the header
files __DECC_INCLUDE_{PRO,EPI}LOGUE.H.
Reviewed-by: Andy Polyakov <appro@openssl.org>
In the case of generating a file like this:
GENERATE[foo.S]=mkfoo.pl arg1 arg2
the 'mkfoo.pl' generator itself might need to include other files,
such as perl modules within our source tree. We can reuse already
existing syntax for it, like this:
INCLUDE[mkfoo.pl]=module/path
or:
DEPEND[mkfoo.pl]=modules/mymodule.pm
This change implements the support for such constructs, and for the
DEPEND statement, for any value that indicates a perl module (.pm
file), it will automatically infer an INCLUDE statement for its
directory, just like it does for C header files, so you won't have do
write this:
DEPEND[mkfoo.pl]=modules/mymodule.pm
INCLUDE[mkfoo.pl]=modules
Reviewed-by: Emilia Käsper <emilia@openssl.org>
The Unix build was the last to retain the classic build scheme. The
new unified scheme has matured enough, even though some details may
need polishing.
Reviewed-by: Rich Salz <rsalz@openssl.org>
This also restores the possibility to have ml used with VC-WIN32 with
no-asm, which was lost during the mk1mf -> unified transition.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Because we have a directory 'test', the target 'test' may be confusing
to make. However, if we make it depend on 'tests', which doesn't
exist, make should never fail to run the actions.
Reviewed-by: Andy Polyakov <appro@openssl.org>
To begin with, the default should have been the import library, not
the DLL itself. However, we don't know what directory it's installed
in either way, so we may as well demand the full path from the user
Reviewed-by: Matt Caswell <matt@openssl.org>
no-stdio does not work with the apps. Since the tests also need the apps
it doesn't support that either. Therefore we disable building of both.
no-autoalginit is not compatible with the apps because it requires explicit
loading of the algorithms, and the apps don't do that. Therefore we disable
building the apps for this option. Similarly the tests depend on the apps
so we also disable the tests. Finally the whole point about no-autoalginit
is to avoid excessive executable sizes when doing static linking. Therefore
we disable "shared" if this option is selected.
Reviewed-by: Richard Levitte <levitte@openssl.org>
The handling was Unix centric, already in Configure. Change that to
just collect the value and let the build file templates figure out
what to do with it.
Reviewed-by: Matt Caswell <matt@openssl.org>
Instead of absolute hard coding of the libz library name, have it use
the macro LIBZ, which is set to defaults we know in case it's
undefined.
This allows our configuration to set something that's sane on current
or older platforms, and allows the user to override it by defining
LIBZ themselves.
Reviewed-by: Matt Caswell <matt@openssl.org>
The macros ZLIB and ZLIB_SHARED weren't appropriately defined,
deviating wrongly from how they worked in earlier OpenSSL versions.
So, restore it so that ZLIB is defined if configured "enable-zlib" and
so that ZLIB and ZLIB_SHARED are defined if configured
"enable-zlib-dynamic".
Additionally, correct the interpretation of the --with-zlib-lib value
on Windows and VMS, where it's used to indicate the actual zlib
zlib library file / logical name, as that can differ depending on zlib
version and packaging on those platforms.
Finally for Windows and VMS, we also define the macro LIBZ with that
file name / logical name when configured "zlib-dynamic", so the
compression unit can pick it up and use it.
Reviewed-by: Matt Caswell <matt@openssl.org>
Since NDEBUG is defined unconditionally on command line for release
builds, we can omit *_DEBUG options in favour of effective "all-on"
in debug builds exercised though CI.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>
Without proper cleanup after DLL link failure second attempt to
run nmake would actually proceed and failure will be "shifted" to
run time. This is because libcrypto.lib import library is generated
even if DLL link fails.
Reviewed-by: Richard Levitte <levitte@openssl.org>
We used to symlink generate_ssl_tests.pl to the build directory.
Now that the build scripts look for sources in both directories, this
is no longer necessary (see commit
fbd361eaf8).
Reviewed-by: Richard Levitte <levitte@openssl.org>
Currently, SSL tests are configured via command-line switches to
ssltest.c. This results in a lot of duplication between ssltest.c and
apps, and a complex setup. ssltest.c is also simply old and needs
maintenance.
Instead, we already have a way to configure SSL servers and clients, so
we leverage that. SSL tests can now be configured from a configuration
file. Test servers and clients are configured using the standard
ssl_conf module. Additional test settings are configured via a test
configuration.
Moreover, since the CONF language involves unnecessary boilerplate, the
test conf itself is generated from a shorter Perl syntax.
The generated testcase files are checked in to the repo to make
it easier to verify that the intended test cases are in fact run; and to
simplify debugging failures.
To demonstrate the approach, min/max protocol tests are converted to the
new format. This change also fixes MinProtocol and MaxProtocol
handling. It was previously requested that an SSL_CTX have both the
server and client flags set for these commands; this clearly can never work.
Guide to this PR:
- test/ssl_test.c - test framework
- test/ssl_test_ctx.* - test configuration structure
- test/handshake_helper.* - new SSL test handshaking code
- test/ssl-tests/ - test configurations
- test/generate_ssl_tests.pl - script for generating CONF-style test
configurations from perl inputs
Reviewed-by: Richard Levitte <levitte@openssl.org>
Previously, we updated the static libraries (libcrypto.a on Unix,
libcrypto.lib on Windows) with all the object files, regardless of if
they were rebuilt or not. With this change, we only update them with
the object files were rebuilt.
NOTE: this does not apply on VMS, as the expansion of $? may be too
large for a command line.
Reviewed-by: Andy Polyakov <appro@openssl.org>
The logic to find out of there are any .d files newer than Makefile is
sound. Checking the result was less so.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reverts commit 087ca80ad8
Instead of battling the odd format of argv given to main() in default
P64 mode, tell the compiler to make it an array of 64-bit pointers
when compiling in P64 mode.
A note is added in NOTES.VMS regarding minimum DEC C version.
Reviewed-by: Andy Polyakov <appro@openssl.org>
- In Configure, register the perl interpreter used to run Configure,
so that's the one being used throughout instead of something else
that Configure happens to find. This is helpful for using a perl
version that's not necessarely first in $PATH:
/opt/perl/5.22.1/bin/perl ./Configure
- Make apps/tsget a generated file, just like apps/CA.pl, so the
perl interpreter registered by Configure becomes the hashbang path
instead of a hardcoded /usr/bin/perl
Reviewed-by: Andy Polyakov <appro@openssl.org>
There are rare cases when an object file will only be used when
building a shared library. To enable this, we introduce
SHARED_SOURCE:
SHARED_SOURCE[libfoo]=dllmain.c
Reviewed-by: Andy Polyakov <appro@openssl.org>
The warning MAYLOSEDATA3 is one you will always get when compiling
source that calculates the difference between two pointers with
/POINTER_SIZE=64.
The reason is quite simple, ptrdiff_t is always a 32-bit integer
regardless of pointer size, so the result of 'ptr1 - ptr2' can
potentially be larger than a 32-bit integer. The compiler simply
warns you of that possibility.
However, we only use pointer difference within objects and strings,
all of them well within 2^32 bytes in size, so that operation is
harmless with our source, and we can therefore safely turn off that
warning.
Reviewed-by: Rich Salz <rsalz@openssl.org>
On Windows and Unix, the staging directory $(DESTDIR) can simply be
prepended to the installation directory. An attempt was made to do
something similar on VMS, but that ended up being a half measure
solution. Instead of that, simply use the staging directory as a
prefix under which [.OPENSSL-INSTALL] and [.OPENSSL-COMMON] will hold
the two directory trees that should end up in the directories
indicated by --prefix and --openssldir, and finish the installation
with appropriate instructions on what to do next.
Reviewed-by: Rich Salz <rsalz@openssl.org>
- on VMS, SYS$COMMON:[SSL] is already used as installation directory
by HP SSL, so we make our default for --openssldir
SYS$COMMON:[OPENSSL-COMMON] instead.
- Updated notes on default installation dirs fir Unix and Windows
Reviewed-by: Rich Salz <rsalz@openssl.org>
Sometimes, you might end up with a rather long compile line due to
excessively long /INCLUDE directories. Compensate for it by making
a temporary logical name with them and using said logical name as
/INCLUDE argument.
A note was added to NOTES.VMS regarding these limitations.
Reviewed-by: Rich Salz <rsalz@openssl.org>
On VMS, we downcase option names, which means that config names are
downcased as well, so they need to be downcased in the target table to
be found.
Reviewed-by: Rich Salz <rsalz@openssl.org>
In constructions such as 'for x in $(MAKEVAR); do ...', there's the
possibility that $(MAKEVAR) is en empty value. Some shells don't like
that, so introduce a dummy value that gets discarded:
for x in dummy $(MAKEVAR); do
if [ "$$x" = "dummy" ]; then continue; fi
Closes RT#4459
Reviewed-by: Rich Salz <rsalz@openssl.org>
utils/mkrc.pl was added a while ago as a better generator for the
Windows DLL resource file. Finalize the change by removing the
ms/version32.rc generator from Configure and adding resource file
support using mkrc.pl in Configurations/windows-makefile.pl
Reviewed-by: Andy Polyakov <appro@openssl.org>
One of the 'generate' targets depended on $(SRCDIR)/apps/progs.h,
which depended on... nothing. This meant it never got regenerated
once it existed, regardless of need. Of course, we could have it
depend on all the files checked to generate it, but they also depend
on progs.h, so we'd end up getting cricular dependencies, which makes
make unhappy.
Furthermore, and this applies for the other generated files, having
them as targets means that they may be regenerated on the fly in some
cases, and since they get written to the source tree, this isn't such
a good idea if that tree is read-only (which is a possible situation
in an out-of-tree build).
So, we move all the actions to the 'generate' targets themselves, thus
making sure they get regenerated in a controlled manner and regardless
of dependencies.
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Some implementations of sed require a newline before an ending '}'.
The easier method is to replace that sed command with the
corresponding perl command.
Closes RT#4448
Reviewed-by: Rich Salz <rsalz@openssl.org>
Depending on what has been built so far, all .d files may not be
present and 'find' will exit with non-zero exit code. This isn't a
bother for us but may break make, so clear the exit code with an added
'exit 0'.
Closes RT#4444
Reviewed-by: Rich Salz <rsalz@openssl.org>
generatesrc() did already receive dependency information, but never
used it, and never really needed to... until now.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Instead of relying on the '-nt' test operator, which doesn't exist
everywhere, use find's '-newer' to find out if any of the known .d
files is newer than Makefile.
Closes RT#4444
Reviewed-by: Andy Polyakov <appro@openssl.org>
On Windows, we set INSTALLTOP to default as follows:
VC-WIN32:
PREFIX: %ProgramFiles(x86)%\OpenSSL
OPENSSLDIR: %CommonProgramFiles(x86)%\SSL
VC-WIN64*:
PREFIX: %ProgramW6432%\OpenSSL
OPENSSLDIR: %CommonProgramW6432%\SSL
Should those environment variables be missing, the following is used
as fallback:
PREFIX: %ProgramFiles%\OpenSSL
OPENSSLDIR: %CommonProgramFiles%\SSL
Reviewed-by: Andy Polyakov <appro@openssl.org>
This is a living document, everyone is encouraged to add to it.
Implementation details as well as broader implementation philosophy
has a place here.
I'm starting with documentation of the how conditions in build.info
files are treated.
Reviewed-by: Rich Salz <rsalz@openssl.org>
In most builds, we can assume that engines live in the build tree
subdirectory "engines". This was hard coded into the tests that use
the engine ossltest.
However, that hard coding is tedious, it would need to be done in
every test recipe, and it's an incorrect assumption in some cases.
This change has us play it safe and let the build files tell the
testing framework where the engines are.
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
util/mkdef.pl assumes it knows what the resulting library name will
be. Really, it shouldn't, but changing it will break classic native
Windows builds, so we leave it for now and change the LIBRARY line
externally when needed instead.
Reviewed-by: Matt Caswell <matt@openssl.org>
Normally we always refer to source files relative to $SRCDIR in Makefiles.
However the reference to unix-Makefile.tmpl was using a fully expanded
absolute path. This can cause problems for Mingw.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
In unix-Makefile.tmpl, this construction has been used a few times
if ! something; then ...
It seems, though, that some shells do not understand !, so these need
to be changed.
Reviewed-by: Rich Salz <rsalz@openssl.org>
When passing down values to Makefile.shared, do so with single quotes
as much as possible to avoid having the shell create a mess of quotes.
Reviewed-by: Rich Salz <rsalz@openssl.org>
The variable SHARED_CFLAGS and SHARD_LDFLAGS were used in the Unix
template because they normally contain options used when building
"shared". The Windows template, on the other hand, uses LIB_CFLAGS,
to express the intended use of those flags rather than their content.
The Windows template still used SHARED_LDFLAGS, which seems
inconsistent.
To harmonize the two, any SHARED_CFLAGS gets renamed to LIB_CFLAGS and
SHARED_LDFLAGS to LIB_LDFLAGS. That makes the intent consistent along
with BIN_{C,LD}FLAGS and DSO_{C,LD}FLAGS.
Finally, make sure to pass down $(LIB_CFLAGS) or $(DSO_CFLAGS) along
with $(CFLAGS) when using Makefile.shared.
Reviewed-by: Rich Salz <rsalz@openssl.org>
It turns out that different sed implementations treat -i differently
to cause issues. make it simpler by avoiding it entirely and give
perl the trust to be consistent enough.
Reviewed-by: Rich Salz <rsalz@openssl.org>
If pre-processor failed, an empty .s file could be left behind,
which could get successfully compiled if one simply re-ran make
and cause linking failures. Not anymore. Remove even intermediate .S
in case of pre-processor failure.
Reviewed-by: Richard Levitte <levitte@openssl.org>
One of them didn't clean away .d.tmp files properly.
The other would overwrite the .d files unconditionally, thereby
causing a possibly unnecessary dependency rebuild, which touches the
date of Makefile, which causes a possibly unnecessary rebuild of
buildinf.h and everything that depends on that.
Reviewed-by: Andy Polyakov <appro@openssl.org>
The source file generators sometimes use $(CC) to post-process
generated source, and getting the inclusion directories may be
necessary at times, so we pass them down.
RT#4406
Reviewed-by: Rich Salz <rsalz@openssl.org>
The reason for this is that the static libraries and the DLL import
libraries are named the same on Windows. When configured "shared",
the static libraries are unused anyway.
Reviewed-by: Rich Salz <rsalz@openssl.org>
This introduces the settings loutflag and aroutflag, because different
Windows tools that do the same thing have different ways to specify
the output file.
The Borland C++ config is commented away for the monent, perhaps
permanently.
Reviewed-by: Rich Salz <rsalz@openssl.org>
This change is a bit more complex, as it involves several recipe
variants.
Also, remove the $(CROSS_COMPILE) prefix for the makedepend program.
When we use the program "makedepend", this doesn't serve anything,
and when we use the compiler, this value isn't even used.
Reviewed-by: Rich Salz <rsalz@openssl.org>
VMS doesn't have "makedepend" anyway, so this is just a matter of using
the right qualifiers when 'makedepend' is enabled.
Reviewed-by: Rich Salz <rsalz@openssl.org>
For assembler, we want the final target to be foo.s (lowercase s).
However, the build.info may have lines like this (note upper case S):
GENERATE[foo.S]=foo.pl
This indicates that foo.s (lowercase s) is still to be produced, but
that producing it will take an extra step via $(CC) -E. Therefore,
the following variants (simplified for display) can be generated:
GENERATE[foo.S]=foo.pl => foo.s: foo.pl
$(PERL) $foo.pl $@.S; \
$(CC) $(CFLAGS) -E -P $@.S > $@ && \
rm -f $@.S
GENERATE[foo.s]=foo.pl => foo.s: foo.pl
$(PERL) $foo.pl $@
GENERATE[foo.S]=foo.m4 => foo.s: foo.m4
m4 -B 8192 $foo.m4 > $@.S; \
$(CC) $(CFLAGS) -E -P $@.S > $@ && \
rm -f $@.S
GENERATE[foo.s]=foo.m4 => foo.s: foo.m4
m4 -B 8192 $foo.m4 > $@
Reviewed-by: Andy Polyakov <appro@openssl.org>
Make all scripts produce .S, make interpretation of $(CFLAGS)
pre-processor's responsibility, start accepting $(PERLASM_SCHEME).
[$(PERLASM_SCHEME) is redundant in this case, because there are
no deviataions between Solaris and Linux assemblers. This is
purely to unify .pl->.S handling across all targets.]
Reviewed-by: Richard Levitte <levitte@openssl.org>
For config targets such as 'dist', which doesn't have a BASE template,
we still need to have a default build scheme.
Additionally, the unified Makefile template's target 'tar' wasn't
quite as flexible as the unixmake one.
Finally, .travis-create-release.sh can be somewhat simplified now that
it builds with the unified build scheme.
Reviewed-by: Matt Caswell <matt@openssl.org>
Should it be needed because the recipes within a RAW section might
clash with those generated by Configure, it's possible to tell it
not to generate them with the use of OVERRIDES, for example:
SOURCE[libfoo]=foo.c bar.c
OVERRIDES=bar.o
BEGINRAW[Makefile(unix)]
bar.o: bar.c
$(CC) $(CFLAGS) -DSPECIAL -c -o $@ $<
ENDRAW[Makefile(unix)]
Reviewed-by: Rich Salz <rsalz@openssl.org>
In some cases, one might want to generate some source files from
others, that's done as follows:
GENERATE[foo.s]=asm/something.pl $(CFLAGS)
GENERATE[bar.s]=asm/bar.S
The value of each GENERATE line is a command line or part of it.
Configure places no rules on the command line, except the the first
item muct be the generator file. It is, however, entirely up to the
build file template to define exactly how those command lines should
be handled, how the output is captured and so on.
Reviewed-by: Rich Salz <rsalz@openssl.org>
When object files with common block symbols are added to static
libraries on Darwin, those symbols are invisible to the linker that
tries to use them. Our solution was to use -fno-common when compiling
C source.
Unfortunately, there is assembler code that defines OPENSSL_ia32cap_P
as a common block symbol, unconditionally, and in some cases, there is
no other definition. -fno-common doesn't help in this case.
However, 'ranlib -c' adds common block symbols to the index of the
static library, which makes them visible to the linker using it, and
that solves the problem we've seen.
The common conclusion is, either use -fno-common or ranlib -c on
Darwin. Since we have common block symbols unconditionally, choosing
the method for our source is easy.
Reviewed-by: Kurt Roeckx <kurt@openssl.org>
We copied $target{cflags}, $target{defines} and a few more to %config,
just to add to the entries. Avoid doing so, and let the build templates
deal with combining the two.
There are a few cases where we still fiddle with %target, but that's
acceptable.
Reviewed-by: Andy Polyakov <appro@openssl.org>
The thread_cflag setting filled a double role, as kinda sorta an
indicator of thread scheme, and as cflags. Some configs also added
lflags and ex_libs for multithreading regardless of if threading would
be enabled or not.
Instead of this, add threading cflags among in the cflag setting,
threading lflags in the lflag setting and so on if and only if threads
are enabled (which they are by default).
Also, for configs where there are no special cflags for threading (the
VMS configs are of that kind), this makes it possible to still clearly
mention what thread scheme is used.
The exact value of thread scheme is currently ignored except when it's
"(unknown)", and thereby only serves as a flag to tell if we know how
to build for multi-threading in a particular config. Yet, the
currently used values are "(unknown)", "pthreads", "uithreads" (a.k.a
solaris threads) and "winthreads".
Reviewed-by: Andy Polyakov <appro@openssl.org>
Instead, make the build type ("debug" or "release") available through
$config{build_type} and let the configs themselves figure out what the
usual settings (such as "cflags", "lflags" and so on) should be
accordingly.
The benefit with this is that we can now have debug and release
variants of any setting, not just those Configure supports, and may
also involve other factors (the MSVC flags /MD[d] and /MT[d] involve
both build type and whether threading is enabled or not)
Reviewed-by: Andy Polyakov <appro@openssl.org>
Configure had the Unix centric addition of -lz when linking with zlib
is enabled, which doesn't work on other platforms. Therefore, we move
it to the BASE_unix config template and add corresponding ones in the
other BASE_* config templates. The Windows one is probably incomplete,
but that doesn't matter for the moment, as mk1mf does it's own thing
anyway.
This required making the %withargs table global, so perl snippets in
the configs can use it.
Reviewed-by: Andy Polyakov <appro@openssl.org>
These BASE templates are intended to hold values that are common for
all configuration variants for whole families of configurations.
So far, three "families" are identified: Unix, Windows and VMS, mostly
characterised by the build system they currently use.
Reviewed-by: Andy Polyakov <appro@openssl.org>
This has no real meaning, except it gives Configure a hint that VC
targets are indeed capable of producing shared objects.
Reviewed-by: Matt Caswell <matt@openssl.org>
There are cases, for example when configuring no-asm, that the added
uplink source files got in the way of the cpuid ones. The best way to
solve this is to separate the two.
Reviewed-by: Andy Polyakov <appro@openssl.org>
When building with GNU C, clang or VMS C, it's more efficient to
generate dependency file and object file in one call rather than two.
Have the dependency output in a temporary file and compare it with the
previous one if available to see if replacement is waranted, thereby
avoiding unnecessary reconstruction of Makefile / descrip.mms.
Github issue #750
Reviewed-by: Rich Salz <rsalz@openssl.org>
I read the PROBLEMS, and they're outdated; nothing I'd put in the
online FAQ, for example. Test-builds work without using these files.
Had to remove the rehash.time stuff from Makefile.in
Reviewed-by: Richard Levitte <levitte@openssl.org>
They now default to " " as separator, but that can be overridden by
having a hash with parameters as last argument. The only currently
recognised parameter is `separator'.
The special separator `undef' will force the result to become a list
rather than a concatenated string.
Reviewed-by: Andy Polyakov <appro@openssl.org>
If the local system doesn't have GNU C or clang, and not even
makedepend, the build will stop because the call of 'makedepend'
fails. This changes so the build won't stop because of such failure.
The result will be empty .d files, and that's ok.
Reviewed-by: Rich Salz <rsalz@openssl.org>
These flags are limitting needlessly, are often patched by packagers,
and should be specified on the configuration command line by anyone
who desires for it to be specific rather than forced by us.
This work was already done with mingw when those configs were worked
on, now it gets applied to the remaining configs.
Reviewed-by: Andy Polyakov <appro@openssl.org>
GNU make will re-exec if (it thinks that) the Makefile has changed.
Just having the target Makefile seems to make it think it has, so we
end up in a look where GNU make re-execs for ever.
The fix is easy, just remove the Makefile target and have the depend
target run the recipe on its own instead of depending on Makefile.
Reviewed-by: Rich Salz <rsalz@openssl.org>
This takes us away from the idea that we know exactly how our static
libraries are going to get used. Instead, we make them available to
build shareable things with, be it other shared libraries or DSOs.
On the other hand, we also have greater control of when the shared
library cflags. They will never be used with object files meant got
binaries, such as apps/openssl or test/test*.
With unified, we take this a bit further and prepare for having to
deal with extra cflags specifically to be used with DSOs (dynamic
engines), libraries and binaries (applications).
Reviewed-by: Rich Salz <rsalz@openssl.org>
So far, MingW shared libraries were named like this
libeay32.dll + libeay32.dll.a
ssleay32.dll + ssleay32.dll.a
That naming scheme is antiquated, a reminicense of SSLeay. We're
therefore changing the scheme to something that's more like the rest
of OpenSSL.
There are two factors to remember:
- Windows libraries have no recorded SOvers, which means that the
shared library version must be encoded in the name. According to
some, it's unwise to encode extra periods in a Windows file name,
so we convert version number periods to underscores.
- MingW has multilib ability. However, DLLs need to reside with the
binaries that use them, so to allow both 32-bit and 64-bit DLLs to
reside in the same place, we add '-x64' in the name of the 64-bit
ones.
The resulting name scheme (for SOver 1.1) is this:
on x86:
libcrypto-1_1.dll + libcrypto.dll.a
libssl-1_1.dll + libssl.dll.a
on x86_64:
libcrypto-1_1-x64.dll + libcrypto.dll.a
libssl-1_1-x64.dll + libssl.dll.a
An observation is that the import lib is the same for both
architectures. Not to worry, though, as they will be installed in
PREFIX/lib/ for x86 and PREFIX/lib64/ for x86_64.
As a side effect, MingW got its own targets in Makefile.shared.
link_dso.mingw-shared and link_app.mingw-shared are aliases for the
corresponding cygwin-shared targets. link_shlib.mingw-shared is,
however, a target separated from the cygwin one.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Better libclean that removes the exact files that have been built,
nothing more and nothing less.
Corrected typo
A couple of editorial changes.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Since we're using the acronym DSO everywhere else and that's a common
name for that kind of object, we might as well do so here as well.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Instead of having the installation recipe rely on special knowledge,
feed it with information, including what shared library files belong
together. For Cygwin and Mingw, that's the .dll and its import
library .dll.a. For Unixen, it's the shared library file name with SO
version and the one without.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Originally, the Makefile.shared targets described what they used as
input for a shared object, be it a shared library or a DSO. It turned
out, however, that the link_o targets were used exclusively for
engines and the link_a targets were for libcrypto and libssl.
This rename fest turns and indication on the kind of input the targets
get to the intention with using them.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Then it can pass around the information where it belongs. The
Makefile templates pick it up along with other target data, the
DSO module gets to pick up the information through
crypto/include/internal/dso_conf.h
Reviewed-by: Andy Polyakov <appro@openssl.org>
- install_sw had a display of text that belongs under the install target
- previous layout installed architecture dependent files in
dev:['prefix'.'arch'.LIB], dev:['prefix'.'arch'.EXE] and
dev:['prefix'.'arch'.ENGINES]. Changed to dev:['prefix'.LIB.'arch'],
dev:['prefix'.EXE.'arch'] and dev:['prefix'.ENGINES.'arch'] instead.
Reviewed-by: Andy Polyakov <appro@openssl.org>
This is done with a simple file name comparison. We could think of
something more elegant in the future.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Adding uplink and applink to some builds was done by "magic", the
configuration for "mingw" only had a macro definition, the Configure
would react to its presence by adding the uplink source files to
cpuid_asm_src, and crypto/build.info inherited dance to get it
compiled, and Makefile.shared made sure applink.o would be
appropriately linked in. That was a lot under the hood.
To replace this, we create a few template configurations in
Configurations/00-base-templates.conf, inherit one of them in the
"mingw" configuration, the rest is just about refering to the
$target{apps_aux_src} / $target{apps_obj} in the right places.
Reviewed-by: Andy Polyakov <appro@openssl.org>
VMS DIFF tries to calculate all the differences, which is slower than
just reading the files and stopping at the first difference. The
latter doesn't exist as a command, so the problem is solved with perl
and File::Compare (has been in core perl since very early version 5).
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
DCL may be in extended parsing style, which makes it less case
insensitive, so when removing a string from another, make sure to get
casing correctly.
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
This isn't the fully featured combination of compiler generated
dependency files and Makefile include directives, but a cheaper
variant of the same.
The dependency files are generated automatically, but then we have the
usual "depend" target. However, we depend on it in the bigger phony
targets that are the most likely to be used. That make this feature
automatic enough.
A side effect is that we can't use the build file's timestamp to check
if reconfiguring might be in order. In its place, we use a flag file
that depends on Configure and the build file template and depend on it
in spots where it makes sense to check for the need to reconfigure.
Reviewed-by: Rich Salz <rsalz@openssl.org>
When cross compiling, we may end up with someting like apps/openssl.exe
and a number of test/*.exe. However, util/shlib_wrap.sh doesn't know
what the executable extension should be, if any, so we need to make
sure it has access to that information when testing, since
OpenSSL::Test uses that script to execute all programs.
Reviewed-by: Rich Salz <rsalz@openssl.org>
All those flags existed because we had all the dependencies versioned
in the repository, and wanted to have it be consistent, no matter what
the local configuration was. Now that the dependencies are gone from
the versioned Makefile.ins, it makes much more sense to use the exact
same flags as when compiling the object files.
Reviewed-by: Rich Salz <rsalz@openssl.org>
.d (.MMS in the VMS world) files with just dependencies are built from
exactly the same conditions as the object files. Therefore, the rules
for them can be built at the same time as the rules for the
corresponding object files.
This removes the requirement for a src2dep function in the build file
templates, and for common.tmpl to call it. In the end, the existence
of depend files is entirely up to the build file.
Reviewed-by: Rich Salz <rsalz@openssl.org>
