How Curie was split up into multiple repositories and how all of these repositories were tested using CI tools before being released.
Posted Thu Sep 21 02:49:08 2017
Go over the plan for Curie v0.002, Curie v0.003, Curie v0.004, MuPDF, and release engineering
Posted Sat May 13 18:00:00 2017
A list of the issues that have been fixed and will need to be fixed before Curie v0.002 is released.
Posted Tue Dec 20 23:57:59 2016

Interface

Review the current one

Zaki: Menu works on branch : project-renard/curie/pull/89

Chirag: Continuing work on the keybindings: project-renard/curie/pull/52

Stan: Status bar: move to another milestone: project-renard/curie/pull/53

Next steps

Stan: Drag and drop project-renard/curie/issues/19

Chirag: documentation project-renard/curie/pull/98

Zaki: Split out general code from the menu PR : project-renard/curie/pull/89

Stan: GTK icon and color theming: project-renard/curie/issues/32

Devops

Vagrant on Travis-CI

Newer GTK+

project-renard/devops/issues/14

Adding support for Mac

Chirag: tried to run on Mac natively (with XQuartz for the display and homebrew to install), but keybindings did not work. Will address with the keybindings pull request.

project-renard/curie/pull/52

Windows installer

https://metacpan.org/pod/Perl::Dist::WiX

http://wixtoolset.org/

https://github.com/project-renard/curie/issues/1#issuecomment-221646498

Posted Sat May 28 02:46:43 2016

In order to build curie on Windows1, we need to build the various Perl dependencies. These Perl dependencies in turn require native libraries such as Gtk+ 3 and MuPDF. Using the MSYS2 package manager, we can install the native libraries and link against them.

The following steps show how to install the dependencies along with how I ended up debugging the build. I'm writing out this post in case the debugging steps are useful months later when I forget how or why I did things.

The following will be assuming 64-bit Windows (x86-64 architecture) throughout.

For those who just want to see the final code, go to the PR here.

Building locally

  1. Install MSYS2. This is a straightforward install into the C:\msys64 directory.

  2. Start the "MinGW-w64 Win64 Shell" from the Start Menu. This shell sets up the proper environment variables needed to use 64-bit libraries for the MinGW64 system.

  3. Use the pacman package manager to update the package database and install the build toolchain.

       pacman -Syu  # update
    
       # Install compiler and build configuration tools
       pacman -S --needed --noconfirm mingw-w64-x86_64-toolchain autoconf automake libtool make patch mingw-w64-x86_64-libtool
    
    

  4. Install the native dependencies.

       pacman -S --needed --noconfirm \
           mingw-w64-x86_64-gobject-introspection \
           mingw-w64-x86_64-cairo \
           mingw-w64-x86_64-gtk3 \
           mingw-w64-x86_64-expat \
           mingw-w64-x86_64-openssl
    
    

  5. Install Perl and cpanm

       pacman -S --needed --noconfirm mingw-w64-x86_64-perl
       yes | cpan App::cpanminus
    
    

    However, we ecounter a problem with this last command.

       Configuring M/MI/MIYAGAWA/App-cpanminus-1.7040.tar.gz with Makefile.PL
       Checking if your kit is complete...
       Looks good
       Generating a dmake-style Makefile
       Writing Makefile for App::cpanminus
       Writing MYMETA.yml and MYMETA.json
         MIYAGAWA/App-cpanminus-1.7040.tar.gz
         C:\msys64\mingw64\bin\perl.exe Makefile.PL -- OK
       Running make for M/MI/MIYAGAWA/App-cpanminus-1.7040.tar.gz
       cp lib/App/cpanminus/fatscript.pm blib\lib\App\cpanminus\fatscript.pm
       cp lib/App/cpanminus.pm blib\lib\App\cpanminus.pm
       "C:\msys64\mingw64\bin\perl.exe" -MExtUtils::Command -e cp -- bin/cpanm blib\script\cpanm
       pl2bat.bat blib\script\cpanm
       'pl2bat.bat' is not recognized as an internal or external command,
       operable program or batch file.
       dmake:  Error code 129, while making 'blib\script\cpanm'
       dmake:  'blib\script\cpanm' removed.
         MIYAGAWA/App-cpanminus-1.7040.tar.gz
         dmake -- NOT OK
       

    The important line here is

       'pl2bat.bat' is not recognized as an internal or external command,
       

    It tries to using pl2bat.bat to install the cpanm script, but it is missing. Let's take a look:

       $ which pl2bat.bat
       which: no pl2bat.bat in (...)
       

    OK, so it is not there. What about without the .bat extension?

       # Where is pl2bat?
       $ which pl2bat
       /mingw64/bin/pl2bat
    
       # What does the beginning of the file look like.
       $ head -1 $(which pl2bat)
       #!perl -w
       

    OK, so it does exist and is a Perl script. We just need it be a Windows batch file.

    Note about pl2bat
    The pl2bat command is a tool that is used on Windows in order to allow for running Perl scripts without having to specify that they must run under Perl. So instead of having to type perl myscript to run code in the myscript file, we can just type myscript and run it as if it was just another executable. This works by wrapping the original code and placing it into a Windows batch file that calls the Perl interpreter on the same file.

    It turns out that MSYS2 has the Perl script pl2bat which does the conversion, but that file was not converted itself! To do that we find the path for the pl2bat file and run pl2bat on it.

       pl2bat $(which pl2bat)
    
    

    We can then install cpanm as before and it works.

       yes | cpan App::cpanminus
    
    

  6. Install the Perl dependencies for curie itself by running the following in a copy of the curie repository:

       cpanm --installdeps .
    
    

    This reads the cpanfile which lists the Perl prerequisites for the project and installs everything needed to build, run, and test the code.

    But we hit an issue here when it attempts to install Gtk3 and in turn its dependency Glib.

       $ cpanm -f Glib
       --> Working on Glib
       Fetching http://www.cpan.org/authors/id/X/XA/XAOC/Glib-1.321.tar.gz ... OK
       Configuring Glib-1.321 ... OK
       Building and testing Glib-1.321 ... ! Installing Glib failed. See C:\msys64\...\build.log for details. Retry with --force to force install it.
       FAIL
       

    Let's take a closer look at what is happening with Glib by adding the --verbose flag:

       $ cpanm --verbose Glib
       # [ redacted configuration output ]
       # [ redacted compilation output ]
       [ LD blib\arch\auto\Glib\Glib.dll ]
       Glib.o:Glib.c:(.text+0x156d): undefined reference to `__imp_glib_major_version'
       Glib.o:Glib.c:(.text+0x15a3): undefined reference to `__imp_glib_minor_version'
       Glib.o:Glib.c:(.text+0x15b1): undefined reference to `__imp_glib_micro_version'
       Glib.o:Glib.c:(.text+0x15e8): undefined reference to `__imp_glib_micro_version'
       Glib.o:Glib.c:(.text+0x1603): undefined reference to `__imp_glib_minor_version'
       Glib.o:Glib.c:(.text+0x160a): undefined reference to `__imp_glib_micro_version'
       GUtils.o:GUtils.c:(.text+0x573): undefined reference to `__imp_glib_micro_version'
       GUtils.o:GUtils.c:(.text+0x583): undefined reference to `__imp_glib_minor_version'
       GUtils.o:GUtils.c:(.text+0x593): undefined reference to `__imp_glib_major_version'
       GParamSpec.o:GParamSpec.c:(.text+0x2ee4): undefined reference to `__imp_g_param_spec_types'
       GParamSpec.o:GParamSpec.c:(.text+0x3032): undefined reference to `__imp_g_param_spec_types'
       GParamSpec.o:GParamSpec.c:(.text+0x3152): undefined reference to `__imp_g_param_spec_types'
       GParamSpec.o:GParamSpec.c:(.text+0x3292): undefined reference to `__imp_g_param_spec_types'
       GParamSpec.o:GParamSpec.c:(.text+0x3347): undefined reference to `__imp_g_param_spec_types'
       GParamSpec.o:GParamSpec.c:(.text+0x3422): more undefined references to `__imp_g_param_spec_types' follow
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.text+0x222e): undefined reference to `ffi_prep_cif'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.text+0x224e): undefined reference to `ffi_call'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.text+0x2607): undefined reference to `ffi_prep_cif'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.text+0x262f): undefined reference to `ffi_call'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.rdata$.refptr.ffi_type_void[.refptr.ffi_type_void]+0x0): undefined reference to `ffi_type_void'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.rdata$.refptr.ffi_type_uint64[.refptr.ffi_type_uint64]+0x0): undefined reference to `ffi_type_uint64'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.rdata$.refptr.ffi_type_sint64[.refptr.ffi_type_sint64]+0x0): undefined reference to `ffi_type_sint64'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.rdata$.refptr.ffi_type_double[.refptr.ffi_type_double]+0x0): undefined reference to `ffi_type_double'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.rdata$.refptr.ffi_type_float[.refptr.ffi_type_float]+0x0): undefined reference to `ffi_type_float'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.rdata$.refptr.ffi_type_pointer[.refptr.ffi_type_pointer]+0x0): undefined reference to `ffi_type_pointer'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.rdata$.refptr.ffi_type_uint32[.refptr.ffi_type_uint32]+0x0): undefined reference to `ffi_type_uint32'
       C:\msys64\mingw64\lib\libgobject-2.0.a(libgobject_2_0_la-gclosure.o):(.rdata$.refptr.ffi_type_sint32[.refptr.ffi_type_sint32]+0x0): undefined reference to `ffi_type_sint32'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gmain.o):(.text+0x25c8): undefined reference to `__imp_timeGetTime'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gmain.o):(.text+0x2608): undefined reference to `__imp_timeGetTime'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x1a6): undefined reference to `pcre_get_stringtable_entries'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x231): undefined reference to `pcre_get_stringnumber'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0xa36): undefined reference to `pcre_exec'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x132f): undefined reference to `pcre_compile2'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x134d): undefined reference to `pcre_fullinfo'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x13c9): undefined reference to `pcre_fullinfo'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x1b44): undefined reference to `pcre_fullinfo'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x1d9c): undefined reference to `pcre_study'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x1e9d): undefined reference to `pcre_config'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x1eb4): undefined reference to `pcre_config'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x1f98): undefined reference to `pcre_fullinfo'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x1fc8): undefined reference to `pcre_fullinfo'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x1ff8): undefined reference to `pcre_fullinfo'
       C:\msys64\mingw64\lib\libglib-2.0.a(libglib_2_0_la-gregex.o):(.text+0x2028): undefined reference to `pcre_fullinfo
       

    OK, that's a lot of errors. Seems that the linker phase (ld) has failed to find several function names in the libglib-2.0.a and libgobject-2.0.a library files.

    Note about library files on Windows

    We see that that library files above both end in the .a extension (for archive). This indicates that this is a static library which means symbols such as function names get resolved at linking time through static linking.

    We can also resolve symbols at runtime through dynamic linking. These files have the extension .dll on Windows.

    Let's take a look at what we expect the linker flags to be for libglib and libgobject by using the pkg-config command:

       # Find the name of the glib and gobject library packages
       $ pkg-config --list-all | grep glib\|gobject
       glib-2.0                            GLib - C Utility Library
       gobject-2.0                         GObject - GLib Type, Object, Parameter and Signal Library
    
       # Get the linker flags
       $ pkg-config --libs glib-2.0 gobject-2.0
       -LC:/msys64/mingw64/lib -lgobject-2.0 -lglib-2.0 -lintl
       

    OK, that looks reasonable. But we see the above errors refer to functions that are prefixed with ffi_ and pcre_. The convention for C libraries is that the prefix referse to the libraries that the functions came from. But we don't see any flags for those libraries (-lffi -lpcre). So when linker attempts static linking against the .a files, it does not know that the other libraries are needed.

    We can try to get the linker flags for static linking by adding the --static flag to pkg-config

       # Get the static linker flags
       $ pkg-config --static --libs glib-2.0 gobject-2.0
       -LC:/msys64/mingw64/lib -LC:/msys64/mingw64/lib/../lib -LC:/msys64/mingw64/lib -lgobject-2.0 -lffi -lglib-2.0 -lintl -pthread -lws2_32 -lole32 -lwinmm -lshlwapi -lpcre -lintl -lpcre
       

    Ah, there we go! Let's compare this to what the Glib module uses when linking:

       # Drop into the Glib installation directory
       $ cpanm --look Glib
    
       # Generate the Makefile
       Glib-1.321 $ perl Makefile.PL verbose
       # [ redacted output ]
       Potential libraries are '-LC:/msys64/mingw64/lib -lgobject-2.0 -lglib-2.0 -lintl -lgthread-2.0 -pthread -lmoldname -lkernel32 -luser32 -lgdi32 -lwinspool -lcomdlg32 -ladvapi32 -lshell32 -lole32 -loleaut32 -lnetapi32 -luuid -lws2_32 -lmpr -lwinmm -lversion -lodbc32 -lodbccp32 -lcomctl32':
       # [ redacted output ]
       Result: "C:\msys64\mingw64\lib\libgobject-2.0.a" "C:\msys64\mingw64\lib\libglib-2.0.a" "C:\msys64\mingw64\lib\libintl.a" "C:\msys64\mingw64\lib\libgthread-2.0.a"
       Generating a dmake-style Makefile
       Writing Makefile for Glib
       Writing MYMETA.yml and MYMETA.json
       

    OK, wait, what's this? We get the list of dynamic linker flags where it says "Potential libraries", but when it says "Result:", a couple of those flags become paths to the corresponding .a files. Let's run dmake with the NOECHO flag to see what is going on:

       Glib-1.321 $ dmake NOECHO=''
       # [ redacted output ]
       [ LD blib\arch\auto\Glib\Glib.dll ]
       g++ Glib.def -o blib\arch\auto\Glib\Glib.dll -mdll -s -L"C:\msys64\mingw64\lib\perl5\core_perl\CORE" -L"C:\msys64\mingw64\lib" Glib.o GError.o GUtils.o GLog.o GType.o GBoxed.o GObject.o GValue.o GClosure.o GSignal.o GMainLoop.o GIOChannel.o GParamSpec.o GKeyFile.o GOption.o GBookmarkFile.o GVariant.o gperl-gtypes.o   "C:\msys64\mingw64\lib\perl5\core_perl\CORE\libperl522.a" "C:\msys64\mingw64\lib\libgobject-2.0.a" "C:\msys64\mingw64\lib\libglib-2.0.a" "C:\msys64\mingw64\lib\libintl.a" "C:\msys64\mingw64\lib\libgthread-2.0.a" -Wl,--enable-auto-image-base
       Glib.o:Glib.c:(.text+0x156d): undefined reference to `__imp_glib_major_version'
       Glib.o:Glib.c:(.text+0x15a3): undefined reference to `__imp_glib_minor_version'
       Glib.o:Glib.c:(.text+0x15b1): undefined reference to `__imp_glib_micro_version'
       # [ rest of the linker errors as before ]
       

    OK, so here we see that all the dynamic linker flags are using the full paths to the .a files. Let's take a look at how Makefile.PL passes on this information to the Makefile

       Glib-1.321 $ grep libgobject-2.0.a Makefile
       EXTRALIBS = "C:\msys64\mingw64\lib\libgobject-2.0.a" "C:\msys64\mingw64\lib\libglib-2.0.a" "C:\msys64\mingw64\lib\libintl.a" "C:\msys64\mingw64\lib\libgthread-2.0.a"
       LDLOADLIBS = "C:\msys64\mingw64\lib\libgobject-2.0.a" "C:\msys64\mingw64\lib\libglib-2.0.a" "C:\msys64\mingw64\lib\libintl.a" "C:\msys64\mingw64\lib\libgthread-2.0.a"
       # EXTRALIBS => q["C:\msys64\mingw64\lib\libgobject-2.0.a" "C:\msys64\mingw64\lib\libglib-2.0.a" "C:\msys64\mingw64\lib\libintl.a" "C:\msys64\mingw64\lib\libgthread-2.0.a"]
       # LDLOADLIBS => q["C:\msys64\mingw64\lib\libgobject-2.0.a" "C:\msys64\mingw64\lib\libglib-2.0.a" "C:\msys64\mingw64\lib\libintl.a" "C:\msys64\mingw64\lib\libgthread-2.0.a"]
       
    ]

    So it appears that ExtUtils::MakeMaker is setting the EXTRALIBS and LDLOADLIBS variables. Let's try setting them ourselves to the output of pkg-config:

       Glib-1.321 $ export GLIB_PKG_CONFIG="$(pkg-config --libs gobject-2.0 glib-2.0)"
       Glib-1.321 $ dmake NOECHO='' LDLOADLIBS="$GLIB_PKG_CONFIG" EXTRALIBS="$GLIB_PKG_CONFIG"
       

    It built successfully! This means that there is a problem with using the full path to the .a files. We need to see why EXTRALIBS and LDLOADLIBS are being set the way they are.

    Let's search the distribution for documentation about these variables by going to the distribution's page on MetaCPAN and typing the query EXTRALIBS LDLOADLIBS in the search box:

    Screenshot of the ExtUtils-MakeMaker page highlighting the Search Distribution field

    Looking at the results, we see ExtUtils::Liblist and ExtUtils::MM_Unix.

    Screenshot of the results of the query for LIB Makefile variables

    The ExtUtils::Liblist looks promising. In version 7.14 of the documentation, there's a section about the Win32 behaviour:

    An entry of the form -lfoo specifies the library foo, which may be spelled differently depending on what kind of compiler you are using. If you are using GCC, it gets translated to libfoo.a, but for other win32 compilers, it becomes foo.lib.

    So the translation to full paths is a part of the design. Reading further, we see

    An entry that matches /:nosearch/i disables all searching for the libraries specified after it.

    So if we but :nosearch in the LIBS parameter of the function that generates the Makefile, (i.e., WriteMakefile). Let's try that

       $ cpanm --look Glib
       Glib-1.321 $ export GLIB_PKG_CONFIG="$(pkg-config --libs gobject-2.0 glib-2.0)"
       Glib-1.321 $ perl Makefile.PL LIBS=":nosearch $GLIB_PKG_CONFIG" verbose
       # [ remove extra output ]
       Potential libraries are ':nosearch -LC:/msys64/mingw64/lib -lgobject-2.0 -lglib-2.0 -lintl -lmoldname -lkernel32 -luser32 -lgdi32 -lwinspool -lcomdlg32 -ladvapi32 -lshell32 -lole32 -loleaut32 -lnetapi32 -luuid -lws2_32 -lmpr -lwinmm -lversion -lodbc32 -lodbccp32 -lcomctl32':
       Result: "-LC:\msys64\mingw64\lib" "-lgobject-2.0" "-lglib-2.0" "-lintl" "-lmoldname" "-lkernel32" "-luser32" "-lgdi32" "-lwinspool" "-lcomdlg32" "-ladvapi32" "-lshell32" "-lole32" "-loleaut32" "-lnetapi32" "-luuid" "-lws2_32" "-lmpr" "-lwinmm" "-lversion" "-lodbc32" "-lodbccp32" "-lcomctl32"
       # [ remove extra output ]
       Glib-1.321 $ dmake NOECHO=''
       

    Hey, it's working!

    So now we know that all we need is to put :nosearch in the LIBS parameter and it will build. But that's still problematic, because if I try to install any module, I will have to open it up to find out what it is passing in to LIBS and copy that into the command line for the configuration step.

    I did that anyway and installed Glib, Cairo, Glib::Object::Introspection, Cairo::GObject, and XML::Parser by manually setting the flags for LIBS.

    I then popped into the #toolchain IRC channel on the irc.perl.org network to share what I had learned and ask if there was another way forward. After a quick convo with mst++, he came back with a one-liner that wrapped the WriteMakefile function and inserted :nosearch into the LIBS parameter:

       perl -Maliased=ExtUtils::MakeMaker,EUMM \
         -e 'my $i = EUMM->can("import"); no warnings "redefine"; *ExtUtils::MakeMaker::import = sub { &$i; my $targ = caller; my $wm = $targ->can("WriteMakefile"); *{"${targ}::WriteMakefile"} = sub { my %args = @_; $args{LIBS} =~ s/^/:nosearch /; $wm->(%args) }; }; do "Makefile.PL" or die "Hack failed: $@"'
       

    Yes, that is one line and it does work!

  7. Rejoice! The curie code runs!

    Screenshot of the curie GUI with a test PDF open

At this point, I was quite happy. Now I needed to reproduce these steps so that they can be used to test every set of changes to the codebase. I could do this with the Appveyor CI service for Windows. Little did I know that there was a lot more work coming my way.

Building on Appveyor

In this section, it is easier to show parts of the Appveyor configuration and explain what each part is for rather than show how I debugged the configuration — mainly because it took me a little over 50 builds on Appveyor to figure out all the quirks of the system. Many of them were missteps on my part, but is hard to figure out where things go wrong on a remote system if you can only run commands every 20 minutes!

  1. After enabling the curie project on the Appveyor website, we need an appveyor.yml configuration file with the commands needed for building the project.

    I first search GitHub for other projects that used Appveyor with MSYS2 and came across the appveyor.yml configuration file for libosmscout. It has support for compiling under both the MinGW64 GCC compiler and Microsoft's MSVC2015 compiler. Right now, I just need the bits for MSYS2 MinGW64:

      version: 1.0.{build}
    
      environment:
        matrix:
        - COMPILER: msys2
          PLATFORM: x64
          MSYS2_ARCH: x86_64
          MSYS2_DIR: msys64
          MSYSTEM: MINGW64
          BIT: 64
    
      install:
        # running under CI
        - set CI_TESTING=1
        - '%APPVEYOR_BUILD_FOLDER%\dev\ci\appveyor\install.bat'
        - 'echo End intall at: & time /t'
    
      build_script:
        - 'echo Nothing to build'
    
      test_script:
        - '%APPVEYOR_BUILD_FOLDER%\dev\ci\appveyor\test.bat'
      

    The important part here is the line

        - '%APPVEYOR_BUILD_FOLDER%\dev\ci\appveyor\install.bat'
      

    which tells Appveyor to run a batch file with more instructions. The contents of this are

      @echo off
    
      cd %APPVEYOR_BUILD_FOLDER%
    
      echo Compiler: %COMPILER%
      echo Architecture: %MSYS2_ARCH%
      echo Platform: %PLATFORM%
      echo MSYS2 directory: %MSYS2_DIR%
      echo MSYS2 system: %MSYSTEM%
      echo Bits: %BIT%
    
      REM Create a writeable TMPDIR
      mkdir %APPVEYOR_BUILD_FOLDER%\tmp
      set TMPDIR=%APPVEYOR_BUILD_FOLDER%\tmp
    
      IF %COMPILER%==msys2 (
        @echo on
        SET "PATH=C:\%MSYS2_DIR%\%MSYSTEM%\bin;C:\%MSYS2_DIR%\usr\bin;%PATH%"
        bash -lc "pacman -S --needed --noconfirm pacman-mirrors"
        bash -lc "pacman -S --needed --noconfirm git"
        REM Update
        bash -lc "pacman -Syu --noconfirm"
    
        REM build tools
        bash -lc "pacman -S --needed --noconfirm mingw-w64-x86_64-toolchain autoconf automake libtool make patch mingw-w64-x86_64-libtool"
    
        REM Set up perl
        bash -lc "pacman -S --needed --noconfirm mingw-w64-x86_64-perl"
        bash -lc "pl2bat $(which pl2bat)"
        bash -lc "yes | cpan App::cpanminus"
        bash -lc "cpanm --notest ExtUtils::MakeMaker"
    
        REM Native deps
        bash -lc "pacman -S --needed --noconfirm mingw-w64-x86_64-gobject-introspection mingw-w64-x86_64-cairo mingw-w64-x86_64-gtk3 mingw-w64-x86_64-expat mingw-w64-x86_64-openssl"
    
        REM There is not a corresponding cc for the mingw64 gcc. So we copy it in place.
        bash -lc "cp -pv /mingw64/bin/gcc /mingw64/bin/cc"
    
        REM Install via cpanfile
        bash -lc "cd $APPVEYOR_BUILD_FOLDER; . $APPVEYOR_BUILD_FOLDER/dev/ci/appveyor/EUMMnosearch.sh; export MAKEFLAGS='-j4 -P4'; cpanm --notest --installdeps ."
      )
    
    

    There is a lot here, but much of it is the same as discussed in the local installation. Let's talk about what is new.

  2. Let's start with the part that installs the dependencies:

      cd $APPVEYOR_BUILD_FOLDER
      . $APPVEYOR_BUILD_FOLDER/dev/ci/appveyor/EUMMnosearch.sh
      export MAKEFLAGS='-j4 -P4'
      cpanm --notest --installdeps .
    
    

    First, we need to be in the directory for the project we want to build (stored in the $APPVEYOR_BUILD_FOLDER environment variable).

    Then we need to set a couple settings so that we can build with the :nosearch setting from before. We do this by sourcing the EUMMnosearch.sh shell script.

  3. The EUMMnosearch.sh script loads that one-liner before by turning that code into a Perl module that is loaded every time that Perl is run. This can be accomplished by setting the PERL5OPT environment variable:

       BUILD_DIR=`(cd $APPVEYOR_BUILD_FOLDER && pwd)`
       export PERL5OPT="-I$BUILD_DIR/dev/ci/appveyor -MEUMMnosearch"
       echo PERL5OPT=$PERL5OPT;
    
    

    This Perl module is only run when the script being run is the configuration step of ExtUtils::MakeMaker:

       package EUMMnosearch;
    
       package main;
       # only run when we call the Makefile.PL script
       if( $0 eq "Makefile.PL"  ) {
           require ExtUtils::MakeMaker;
    
           my $i = ExtUtils::MakeMaker->can("import");
           no warnings "redefine";
           *ExtUtils::MakeMaker::import = sub {
               &$i;
               my $targ = "main";
               my $wm = $targ->can("WriteMakefile");
               *{"${targ}::WriteMakefile"} = sub {
                   my %args = @_;
                   # Only apply :nosearch after lib linker directory
                   # for entire mingw64 system. This way XS modules
                   # that depend on other XS modules can compile
                   # statically using .a files.
                   $args{LIBS} =~ s,^(.*?)(\Q-LC:/msys64/mingw64/lib\E\s),$1 :nosearch $2,;
    
                   # Special case for expat (XML::Parser::Expat) because
                   # it does not use either of
                   #
                   #   - -L<libpath>
                   #   - pkg-config --libs expat
                   $args{LIBS} =~ s,(\Q-lexpat\E),:nosearch $1,;
                   print "LIBS: $args{LIBS}\n";
                   $wm->(%args);
               };
           };
    
           do "Makefile.PL" or die "Hack failed: $@";
    
           # we can exit now that we are done
           exit 0;
       }
    
       1;
    
    

    There are a couple changes to the function that overrides WriteMakefile.

    First, we no longer stick :nosearch in front of all flags in LIBS. Instead, we only stick it before we set the linker search path to the C:/msys64/mingw64/lib directory. That way if there are some non-system libraries that need to be compiled statically, they can still be compiled statically. This happens in the case of Cairo::GObject where the code needs to be linked against the compiled Perl code (called XS) for the Cairo and GObject libraries so that it can access some helper functions provided in those .a files. The rest of the linker flags need to be compiled dynamically so we apply :nosearch to only the reset.

    Second, we insert a special case so that we can build XML::Parser::Expat. Since the LIBS parameter is just set to "-lexpat" in that module, there is no -LC:/msys64/mingw64/lib flag that we can use to insert :nosearch. So we insert it just before to set LIBS to ":nosearch -lexpat" instead.

  4. This all would work on a clean install of Windows with only MSYS2 installed. However, Appveyor is not quite that.

    I came across two more build errors that had nothing to do with MSYS2:

    First, I had tried to install Net::SSLeay module which depends on the libopenssl library. When I tried to install it, I got errors about missing symbols. Everything else was linking properly, so I didn't get why it would not link. I added a line to show me the libraries that the build was using for libopenssl.

       $ cpanm --verbose --configure-args verbose Net::SSLeay
       Configuring Net-SSLeay-1.74 ... WARNING: can't open config file: /usr/local/ssl/openssl.cnf
       *** Found OpenSSL-1.0.2e installed in C:OpenSSL-Win32
       *** Be sure to use the same compiler and options to compile your OpenSSL, perl,
           and Net::SSLeay. Mixing and matching compilers is not supported.
       

    It's looking for the SSL library in C:OpenSSL-Win32. That's not a MinGW64 path! I looked around in the Net::SSLeay code and saw that there was README.Win32 file which says

    2. If your OpenSSL is installed in an unusual place, you can tell Net-SSLeay where to find it with the OPENSSL_PREFIX environment variable: set OPENSSL_PREFIX=c:OpenSSL-1.0.1c perl Makefile.PL make .....

    In our case, we can use

       export OPENSSL_PREFIX="/c/msys64/mingw64"
    
    

    and Net::SSLeay builds properly.

  5. Another issue that I hit was when I tried to install the Alien::MuPDF dependency. This is a module that I wrote that downloads and builds the MuPDF library. I was getting a very odd error from the linker in this case:

       $ cpanm --verbose Alien::MuPDF
       # [ remove extra build output ]
           CC build/release/platform/x11/pdfapp.o
       windres platform/x11/win_res.rc build/release/platform/x11/win_res.o
           LINK build/release/mupdf
       build/release/platform/x11/win_main.o:win_main.c:(.text+0x2d79): undefined reference to `fz_argv_from_wargv'
       build/release/platform/x11/win_main.o:win_main.c:(.text+0x2d79): relocation truncated to fit: R_X86_64_PC32 against undefined symbol `fz_argv_from_wargv'
       build/release/platform/x11/win_main.o:win_main.c:(.text+0x3174): undefined reference to `fz_free_argv'
       build/release/platform/x11/win_main.o:win_main.c:(.text+0x3174): relocation truncated to fit: R_X86_64_PC32 against undefined symbol `fz_free_argv'
       collect2: error: ld returned 1 exit status
       Makefile:303: recipe for target 'build/release/mupdf' failed
       make: *** [build/release/mupdf] Error 1
       External command (make HAVE_GLFW=no) failed! Error: 512
        at ./Build line 58.
       Failed
       Build not completed at ./Build line 58.
       ! Installing Alien::MuPDF failed. See C:\msys64\home\appveyor\.cpanm\work\1462215039.2780\build.log for details. Retry with --force to force install it.
       Searching Modern::Perl (0) on cpanmetadb ...
       FAIL
       

    The fz_ prefix is used for functions in the MuPDF library (also known as libfitz). This time there are undefined references to files within the same project! Very curious!

    Since I was already suspicious that I could not trust the Appveyor environment, I guessed that this might have to do with differences between the compiler and the linker used.

    To debug this, I tried to build the MuPDF library locally and noticed that the C compiler used is cc. On properly set up systems, cc is usually an alias for a specific C compiler such as gcc. To check this, I ran the following on Appveyor:

       # Information about the toolchain
       $ which -a cc; cc -v
       /usr/bin/cc
       /bin/cc
       /usr/bin/cc
       Using built-in specs.
       COLLECT_GCC=cc
       COLLECT_LTO_WRAPPER=/usr/lib/gcc/x86_64-pc-msys/4.9.2/lto-wrapper.exe
       Target: x86_64-pc-msys
       Configured with: /msys_scripts/gcc/src/gcc-4.9.2/configure --build=x86_64-pc-msys --prefix=/usr --libexecdir=/usr/lib --enable-bootstrap --enable-shared --enable-shared-libgcc --enable-static --enable-version-specific-runtime-libs --with-arch=x86-64 --disable-multilib --with-tune=generic --enable-__cxa_atexit --with-dwarf2 --enable-languages=c,c++,fortran,lto --enable-graphite --enable-threads=posix --enable-libatomic --enable-libgomp --disable-libitm --enable-libquadmath --enable-libquadmath-support --enable-libssp --disable-win32-registry --disable-symvers --with-gnu-ld --with-gnu-as --disable-isl-version-check --enable-checking=release --without-libiconv-prefix --without-libintl-prefix --with-system-zlib
       Thread model: posix
       gcc version 4.9.2 (GCC)
    
       $ which -a gcc; gcc -v
       /mingw64/bin/gcc
       /usr/bin/gcc
       /bin/gcc
       /MINGW64/bin/gcc
       /usr/bin/gcc
       Using built-in specs.
       COLLECT_GCC=C:\msys64\mingw64\bin\gcc.exe
       COLLECT_LTO_WRAPPER=C:/msys64/mingw64/bin/../lib/gcc/x86_64-w64-mingw32/5.3.0/lto-wrapper.exe
       Target: x86_64-w64-mingw32
       Configured with: ../gcc-5.3.0/configure --prefix=/mingw64 --with-local-prefix=/mingw64/local --build=x86_64-w64-mingw32 --host=x86_64-w64-mingw32 --target=x86_64-w64-mingw32 --with-native-system-header-dir=/mingw64/x86_64-w64-mingw32/include --libexecdir=/mingw64/lib --with-gxx-include-dir=/mingw64/include/c++/5.3.0 --enable-bootstrap --with-arch=x86-64 --with-tune=generic --enable-languages=c,lto,c++,objc,obj-c++,fortran,ada --enable-shared --enable-static --enable-libatomic --enable-threads=posix --enable-graphite --enable-fully-dynamic-string --enable-libstdcxx-time=yes --disable-libstdcxx-pch --disable-libstdcxx-debug --enable-version-specific-runtime-libs --disable-isl-version-check --enable-lto --enable-libgomp --disable-multilib --enable-checking=release --disable-rpath --disable-win32-registry --disable-nls --disable-werror --disable-symvers --with-libiconv --with-system-zlib --with-gmp=/mingw64 --with-mpfr=/mingw64 --with-mpc=/mingw64 --with-isl=/mingw64 --with-pkgversion='Rev1, Built by MSYS2 project' --with-bugurl=http://sourceforge.net/projects/msys2 --with-gnu-as --with-gnu-ld
       Thread model: posix
       gcc version 5.3.0 (Rev1, Built by MSYS2 project)
    
       $ which -a ld; ld -V
       /mingw64/bin/ld
       /usr/bin/ld
       /bin/ld
       /MINGW64/bin/ld
       /usr/bin/ld
       GNU ld (GNU Binutils) 2.25.1
         Supported emulations:
          i386pep
          i386pe
       

    Here we see that both ld and gcc are found under the /mingw64/bin, but cc is found elsewhere and is a different version altogether. The fix for this is simple, just copy gcc to cc

       cp -pv /mingw64/bin/gcc /mingw64/bin/cc
    
    

    and then Alien::MuPDF compiles properly.

  6. Write a post documenting your travails. Happily use Appveyor to test all your code on Windows.

The full pull request for this Appveyor configuration is available here.


  1. Portability is important to writing robust software and eventually curie will have an installer of Windows, so it is best to address any issues early through continuous integration. ↩

Posted Tue May 3 03:47:24 2016

Chirag

  • Work on the Vagrant setup: choose a better base box project-renard/devops/issues/4

  • Refactor tests project-renard/curie/issues/58

  • Add the final tests for the keybinding PR project-renard/curie/pull/52

    (how to to simulate key press events: project-renard/curie/issues/64 )

Zaki

  • Embed AbiWord project-renard/curie/issues/57

  • Embed Evince project-renard/curie/issues/56

Posted Sat Apr 2 16:47:49 2016
Posted Sun Mar 27 18:16:04 2016

Tasks

Zaki

  • Create a table of Evince and Okular menus on the wiki
  • Write GUI tests using mock document and xvfb project-renard/curie/issues/35 (PR at project-renard/curie/pull/41 )

Stan

  • Show number of pages project-renard/curie/issues/21 (PR at project-renard/curie/pull/43 ).
GUI showing the label with the number of pages in a PDF document

and the corresponding unit tests (issue: project-renard/curie/issues/45 , PR: project-renard/curie/pull/51 )

Chirag

  • Deactivate navigation buttons at start and end of document project-renard/curie/issues/22 (PR at project-renard/curie/pull/37 ).
Disable navigation buttons on first page
Disable navigation buttons on last page

Zaki

  • Refactor Chirag's pull request (issue at project-renard/curie/issues/38 and PR at project-renard/curie/pull/39 ).

  • Check for errors using Perl::Critic and Test::Spelling. This can be used to check for irregular whitespace and spelling errors among other things (issue at project-renard/curie/issues/46 and PR at project-renard/curie/pull/47 )

Travis-CI: Shows whitespace and code stricture failures
Travis-CI: Fix the stricture failures, but whitespace failures remain
Travis-CI: Shows documentation spelling failure
Travis-CI: Shows all tests (Perl::Critic and POD spelling) now passing
  • Add a compatibility layer for GtkScrolledWindow when using a Gtk+ version less v3.8 (issue at project-renard/curie/issues/42 and PR at project-renard/curie/pull/48 )
Posted Sat Mar 19 20:29:28 2016

We need to start testing the code and automatically running those tests using continuous integration.

Looked at a review of PDF metadata extraction doc (mirrored here) from this blog.

Adding Travis-CI support: - Go to https://travis-ci.org/, select the + sign, click on project-renard https://travis-ci.org/profile/project-renard

  • Zaki: need to add gir1.2-gdl-3 package for docking UIs to Travis-CI package whitelist (issues are at project-renard/curie/issues/34 , travis-ci/apt-package-whitelist/issues/2700 , travis-ci/apt-package-whitelist/issues/2702 ).

Issues carried over from last week:

  • Stan: Show number of pages project-renard/curie/issues/21

  • Chirag: Deactivate navigation buttons at start and end of document project-renard/curie/issues/22

Posted Sat Mar 12 01:12:49 2016

The pull requests from the previous meeting were discussed and merged. Since all of the work was performed on a single file, it is now necessary to start splitting off the code into different modules so that it is easier to work with.

Zaki: PR to refactor the script project-renard/curie/pull/24

curie GUI with page number entry box and scrollbar
  • Stan: Show number of pages project-renard/curie/issues/21

  • Chirag: Deactivate navigation buttons at start and end of document project-renard/curie/issues/22

  • Zaki: Work with PNG data in-memory: PR to create Cairo::ImageSurface in-memory project-renard/curie/pull/30 which addresses issue project-renard/curie/issues/25

Posted Sun Mar 6 21:30:50 2016