Vulkan-Docs/doc/specs/vulkan

= Vulkan^(R)^ Specification Build Instructions and Notes
:toc2:
:toclevels: 1


[[intro]]
== Introduction

This README describes important stuff for getting the Vulkan API
specification and reference pages building properly.


[[building]]
== Building The Spec

Once you have all the right tools installed (see <<depends,Software
Dependencies>> below), go to `...path-to-git-repo/doc/specs/vulkan` .

    $ make all

or make the individual targets `html`, `pdf`, `styleguide`, `manhtml`,
`manpdf`, `manhtmlpages`, `checkinc`, and `checklinks`.

NOTE: The `validusage` target is not built as part of `make all`, due to it
needing to be built with all extensions - it will fail without.

These targets generate a variety of output documents in the directory
specified by the Makefile variable `$(OUTDIR)` (by default,
`../../../out/1.0`).
The checked-in file `../../../out/1.0/index.html` links to all these
targets, or they can individually be found as follows:

  * API spec:
  ** `html` - Single-file HTML5 in `$(OUTDIR)/html/vkspec.html`
  ** `pdf` - PDF in `$(OUTDIR)/pdf/vkspec.pdf`
  * "`Vulkan Documentation and Extensions`" guide:
  ** `styleguide` - Single-file HTML5 in `$(OUTDIR)/styleguide.html`
  * Diff spec:
  ** `diff_html` - Single-file HTML5 in `$(OUTDIR)/html/diff.html`
  * Reference pages:
  ** `manhtml` - Single-file HTML in `$(OUTDIR)/apispec.html`
  ** `manpdf` - Single-file PDF in `$(OUTDIR)/apispec.html`
  ** `manhtmlpages` - File-per-entry-point HTML in `$(OUTDIR)/man/html/*`
  * Validator output:
  ** `checkinc` - List of commands, structs, etc.
     missing from the API spec in `$(OUTDIR)/checks/notInSpec.txt`
  ** `checklinks` - Validator script output for API spec in
     `$(OUTDIR)/checks/specErrs.txt` and for reference pages in
     `$(OUTDIR)/checks/manErrs.txt`
  * Valid usage database:
  ** `validusage` - json database of all valid usage statements in the
     specification. Must be built with ./makeAllExts (for now).
     Output in `$(OUTDIR)/validation/validusage.json`.
     A validated schema for the output of this is stored in
     `$(CURDIR)/config/vu-to-json/vu_schema.json`

Once you have the basic build working, an appropriate parallelization option
to make, such as

----
make -j 8
----

may significantly speed up the reference page builds.


[[build-bugs]]
=== Asciidoctor Build Errors

If you see an error like this from the `pdf` target:

    /home/jon/.rbenv/versions/2.3.3/lib/ruby/gems/2.3.0/gems/ruby-enum-0.7.1/lib/ruby-enum/enum.rb:34:in `const_set': asciidoctor: FAILED: /home/tree/git/vulkan/doc/specs/vulkan/vkspec.txt: Failed to load AsciiDoc document - wrong constant name default (NameError)

then try <<ruby-enum-downgrade,downgrading ruby-enum>>
as described below


[[building-extensions]]
=== Building With Extensions Included

Extensions are defined in the same source as the core Specification, but
are only conditionally included in the output.
Asciidoctor http://asciidoctor.org/docs/user-manual/#attributes[attributes]
of the same name as the extension are used to define whether the extension
is included or not - defining such an attribute will cause the output to
include the text for that extension.

When building the specification, the extensions included are those specified
as a space-separated list of extension names (e.g. `VK_KHR_surface`) in the
Makefile variable `$(EXTENSIONS)`, normally set on the make command line.
When changing the list of extensions, it is critical to remove all generated
files using the `clean_generated` Makefile target, as the contents of
generated files depends on `$(EXTENSIONS)`.
There are several helper scripts which clean these files and then build one
or more specified targets for specified extensions:

  * `makeExt` - generate outputs with a single extension enabled.
    Usage is `makeExt extension-name target(s)`, where `extension-name` is
    the extension name string, such as `VK_EXT_debug_report`.
  * `makeKHR` - generate outputs with all Khronos (`VK_KHR_*`) extensions
    enabled.
    Usage is `makeKHR target(s)`.
  * `makeKHRAndKHX` - generate outputs with all Khronos (`VK_KHR_*`) and
    Khronox Experimental (`VK_KHX_*`) extensions enabled.
    Usage is `makeKHRAndKHX target(s)`.

The Makefile variable `$(APITITLE)` defines an additional string which is
appended to the specification title.
When building with extensions enabled, this should be set to something like
`(with extension VK_extension_name)`.
The `makeExt`, `makeKHR`, and `makeKHRAndKHX` scripts already do this.


[[building-diff]]
==== Building A Highlighted Extension Diff

The "diff_html" target in the makefile can be used to generate a version of
the specification which highlights changes made to the specification by the
inclusion of a particular set of extensions.

Extensions in the "EXTENSIONS" attribute defines the base extensions to be
enabled by the specification, and these will not be highlighted in the
output.
Extensions in the "DIFFEXTENSIONS" attribute defines the set of extensions
whose changes to the text will be highlighted when they are enabled.
Any extensions in both environment variables will be treated as if they were
only included in DIFFEXTENSIONS.
The DIFFEXTENSIONS environment variable can be used alongside the make*
scripts in this repository.

In the resulting html document, content that has been added by one of the
extensions will be highlighted with a lime background, and content that was
removed will be highlighted with a pink background.
Each section has an anchor of "#differenceN", with an arrow (=>) at the end
of each section which links to the next difference section.
The first diff section is "difference1".


[[building-test]]
=== Alternate and Test Builds

If you are just testing asciidoc formatting, macros, stylesheets, etc., you
may want to edit `vkspec.txt` to just include your test code.
The asciidoctor HTML build is very fast, even for the whole Specification,
but PDF builds take several minutes.


=== Rebuilding The Generated Images

There are some images in the `images/` directory which are maintained in one
format but need to be converted to another format for corresponding types of
output.
Most are SVG converted to PDF, some are PPT converted to PDF converted to
SVG.
SVG are needed by all builds.

These files are not automatically converted by the Makefile.
Instead, all output forms required are checked into `images/` .
On the rare occasions that someone changes a source document and needs to
regenerate the other forms:

----
cd images ; make
----


=== Validation Scripts

There are a several Makefile targets which look for inconsistencies and
missing material between the specification and ref pages, and the canonical
description of the API in `vk.xml` :

  * `checkinc`
  * `checklinks`
  * `allchecks` - both `checkinc` and `checklinks`

They are necessarily heuristic since they're dealing with lots of
hand-written material.
To use them you'll also need to install:

  * `python3`

The `checkinc` target uses Unix filters to determine which autogenerated API
include files are used (and not used) in the spec.
It generates several output files, but the only one you're likely to care
about is `actual.only`.
This is a list of the include files which are *not* referenced anywhere in
the spec, and probably correspond to undocumented material in the spec.

The `checklinks` target validates the various internal tagged links in the
man pages and spec (e.g. the `fname:vkFuncBlah`, `sname:VkStructBlah`, etc.)
against the canonical description of the API in `vk.xml`.
It generates two output files, `manErrs.txt` and `specErrs.txt`, which
report problematic tags and the filenames/lines on which those tags were
found.




[[macros]]
== Our Asciidoc Macros

We use a bunch of custom macros in the reference pages and API spec asciidoc
sources.
The validator scripts rely on these macros as part of their sanity checks,
and you should use the macros whenever referring to an API command, struct,
token, or enum name, so the documents are semantically tagged and more
easily verifiable.

The supported macros are defined in the `config/vulkan-macros/extension.rb`
asciidoctor extension script.

The tags used are described in the style guide (`styleguide.txt`).

We (may) eventually tool up the spec and ref pages to the point that
anywhere there's a type or token referred to, clicking on (or perhaps
hovering over) it in the HTML view and be taken to the definition of that
type/token.
That will take some more plumbing work to tag the stuff in the autogenerated
include files, and do something sensible in the spec (e.g. resolve links to
internal references).

Most of these macros deeply need more intuitive names.


[[refpages]]
== Reference Pages

The reference pages are extracted from the API Specification source, which
has been tagged to help identify boundaries of language talking about
different commands, structures, enumerants, and other types.
A set of Python scripts extract and lightly massage the relevant tagged
language into corresponding ref page.
Pages without corresponding content in the API spec are generated
automatically, when possible (e.g. for `Vk*FlagBits` pages).

If for some reason you want to regenerate the ref pages from scratch
yourself, you can do so by

----
rm man/apispec.txt
make apispec.txt
----

The `genRef.py` script will generate many warnings, but most are just
reminders that some pages are automatically generated.
If everything is working correctly, all the `man/*.txt` files will be
regenerated, but their contents will not change.

If you add new API features to the Specification in a branch, make sure that
the commands have the required tagging and that ref pages are generated for
them, and build properly.


[[styles]]
== Our stylesheets

NOTE: Section mostly TBD.

We use the default Asciidoctor stylesheet.


=== Marking Normative Language

Normative language is marked as *bold*, and also with the [purple]#purple#
role for html output.
It can be used to mark entire paragraphs or spans of words.
In addition, the normative terminology macros, such as must: and may: and
cannot:, always use this role.

The formatting of normative language depends on the stylesheet.
Currently it just comes out in purple.
We may add a way to disable this formatting at build time.


[[equations]]
== Imbedding Equations

Where possible, equations should be written using straight asciidoc markup
using the _eq_ role.
This covers many common equations and is faster than the alternatives.

For more complex equations, such as multi-case statements, matrices, and
complex fractions, equations should be written using the latexmath: inline
and block macros.
The contents of the latexmath: blocks should be LaTeX math notation.
LaTeX math markup delimiters are now inserted by the asciidoctor toolchain.

LaTeX math is passed through unmodified to all HTML output forms, which is
subsequently rendered with the KaTeX engine when the html is loaded.
A local copy of the KaTeX release is kept in `doc/specs/vulkan/katex` and
copied to the HTML output directory during spec generation.
Math is processed into SVGs via asciidoctor-mathematical for PDF output.

The following caveats apply:

  * The special characters `<` , `>` , and `&` can currently be used only in
    +++[latexmath]+++ block macros, not in +++latexmath:[]+++ inline macros.
    Instead use `\lt`, `\leq`, `\gt`, and `\geq` for `<`, `<=`, `>`, and
    `>=` respectively.
    `&` is an alignment construct for multiline equations, and should only
    appear in block macros anyway.
  * AMSmath environments (e.g. pass:[\begin{equation*}], pass:[{align*}],
    etc.) cannot be used in KaTeX at present, and have been replaced with
    constructs supported by KaTeX such as pass:[{aligned}].
  * Arbitrary LaTeX constructs cannot be used.
    KaTeX and asciidoctor-mathematical are only equation renderers, not full
    LaTeX engines.
    Imbedding LaTeX like \Large or pass:[\hbox{\tt\small VK\_FOO}] may not
    work in any of the backends, and should be avoided.

See the "`Vulkan Documentation and Extensions`" document for more details of
supported LaTeX math constructs.


[[anchors]]
== Asciidoc Anchors And Xrefs

In the API spec, sections can have anchors (labels) applied with the
following syntax.
In general the anchor should immediately precede the chapter or section
title and should use the form '+++[[chapter-section-label]]+++'.
For example,

For example, in chapter +synchronization.txt+:

----
[[synchronization-primitives]]
Synchronization Primitives
----

Cross-references to those anchors can then be generated with, for example,

----
See the <<synchronization-primitives>> section for discussion of fences,
semaphores, and events.
----

You can also add anchors on arbitrary paragraphs, using a similar naming
scheme.

Anything whose definition comes from one of the autogenerated API include
files (`.txt` files in the directories `basetypes`, `enums`, `flags`,
`funcpointers`, `handles`, `protos`, and `structs`) has a corresponding
anchor whose name is the name of the function, struct, etc.
being defined.
Therefore you can say something like:

----
Fences are used with the +++<<vkQueueSubmit>>+++ command...
----


[[depends]]
== Software Dependencies

This section describes the software components used by the Vulkan spec
toolchain.

Before building the Vulkan spec, you must install the following tools:

  * GNU make (make version: 4.0.8-1; older versions probably OK)
  * Python 3 (python, version: 3.4.2)
  * Ruby (ruby, version: 2.3.3)
  ** The Ruby development package (ruby-dev) may also be required in some
     environments.
  * Git command-line client (git, version: 2.1.4).
    The build can progress without a git client, but branch/commit
    information will be omitted from the build.
    Any version supporting the following operations should work:
  ** `git symbolic-ref --short HEAD`
  ** `git log -1 --format="%H"`
  * Ghostscript (ghostscript, version: 9.10).
    This is for the PDF build, and it can still progress without it.
    Ghostscript is used to optimize the size of the PDF, so will be a lot
    smaller if it is included.

The following Ruby Gems and platform package dependencies must also be
installed.
This process is described in more detail for individual platforms and
environment managers below.
Please read the remainder of this document (other than platform-specific
parts you don't use) completely before trying to install.

  * Asciidoctor (asciidoctor, version: 1.5.5)
  * Coderay (coderay, version 1.1.1)
  * JSON Schema (json-schema, version 2.0.0)
  * Asciidoctor PDF (asciidoctor-pdf, version: 1.5.0.alpha15)
  * Asciidoctor Mathematical (asciidoctor-mathematical, version 0.2.2)
  * https://github.com/asciidoctor/asciidoctor-mathematical#dependencies[Dependencies
    for asciidoctor-mathematical] (There are a lot of these!)
  * KaTeX distribution (version 0.7.0 from https://github.com/Khan/KaTeX .
    This is cached under `doc/specs/vulkan/katex/`, and need not be
    installed from github.

.Note
[NOTE]
====
Asciidoctor-pdf versions before `1.5.0.alpha15` have issues with multi-page
valid usage blocks, in that the background only renders for the first page.
`alpha.15` fixes this issue (as well as a few others); do not use prior
versions.
====

Only the `asciidoctor` and `coderay` gems are needed if you don't intend to
build PDF versions of the spec and supporting documents.

`json-schema` is only required in order to validate the output of the valid
usage extraction scripts to a JSON file.
If not installed, validation will be skipped when the JSON is built.

[NOTE]
.Note
====
While it's easier to install just the toolchain components for HTML builds,
people submitting MRs with substantial changes to the Specification are
responsible for verifying that their branches build *both* `html` and `pdf`
targets.
====

Platform-specific toolchain instructions follow:

  * Microsoft Windows
  ** <<depends-ubuntu, Ubuntu / Windows 10>>
  ** <<depends-mingw,MinGW>> (PDF builds not tested)
  ** <<depends-cygwin, Cygwin>>
  * <<depends-osx,Mac OS X>>
  * <<depends-linux,Linux (Debian, Ubuntu, etc.)>>


[[depends-windows]]
=== Windows (General)

Most of the dependencies on Linux packages are light enough that it's
possible to build the spec natively in Windows, but it means bypassing the
makefile and calling functions directly.
This might be solved in future.
For now, there are three options for Windows users: Ubuntu / Windows 10,
MinGW, or Cygwin.


[[depends-ubuntu]]
==== Ubuntu / Windows 10

When using the "`Ubuntu Subsystem`" for Windows 10, most dependencies can be
installed via apt-get:

----
sudo apt-get -qq -y install build-essential python3 git cmake bison flex \
    libffi-dev libxml2-dev libgdk-pixbuf2.0-dev libcairo2-dev \
    libpango1.0-dev ttf-lyx gtk-doc-tools ghostscript
----

The default ruby packages on Ubuntu are fairly out of date.
Ubuntu only provides `ruby` and `ruby2.0` - the latter is multiple revisions
behind the current stable branch, and would require wrangling to get the
makefile working with it.

Luckily, there are better options; either https://rvm.io[rvm] or
https://github.com/rbenv/rbenv[rbenv] is recommended to install a more
recent version.

[NOTE]
.Note
====

  * If you are new to Ruby, you should *completely remove* (through the
    package manager, e.g. `sudo apt-get remove *packagename*`) all existing
    Ruby and asciidoctor infrastructure on your machine before trying to use
    rvm or rbenv for the first time.
    `dpkg -l | egrep 'asciidoctor|ruby|rbenv|rvm'` will give you a list of
    candidate package names to remove.
  ** If you already have a favorite Ruby package manager, ignore this
     advice, and just install the required OS packages and gems.
  * In addition, `rvm` and `rbenv` are *mutually incompatible*.
    They both rely on inserting shims and `$PATH` modifications in your bash
    shell.
    If you already have one of these installed and are familiar with it,
    it's probably best to stay with that one.
    One of the editors, who is new to Ruby, found `rbenv` far more
    comprehensible than `rvm`.
    The other editor likes `rvm` better.
  ** Neither `rvm` nor `rbenv` work, out of the box, when invoked from
     non-Bash shells like tcsh.
     This can be hacked up by setting the right environment variables and
     PATH additions based on a bash environment.
  * Most of the tools on Bash for Windows are quite happy with Windows line
    endings (CR LF), but bash scripts expect Unix line endings (LF).
    The file `.gitattributes` at the top of the vulkan tree in the 1.0
    branch forces such scripts to be checked out with the proper line
    endings on non-Linux platforms.
    If you add new scripts whose names don't end in `.sh`, they should be
    included in .gitattributes as well.
====


[[depends-ubuntu-rbenv]]
===== Ubuntu/Windows 10 Using Rbenv

Rbenv is a lighter-weight Ruby environment manager with less functionality
than rvm.
Its primary task is to manage different Ruby versions, while rvm has
additional functionality such as managing "`gemsets`" that is irrelevant to
our needs.

A complete installation script for the toolchain on Ubuntu for Windows,
developed on an essentially out-of-the-box environment, follows.
If you try this, don't try to execute the entire thing at once.
Do each step separately in case of errors we didn't encounter.

----
# Install packages needed by `ruby_build` and by toolchain components.
# See https://github.com/rbenv/ruby-build/wiki and
# https://github.com/asciidoctor/asciidoctor-mathematical#dependencies

sudo apt-get install autoconf bison build-essential libssl-dev \
    libyaml-dev libreadline6-dev zlib1g-dev libncurses5-dev \
    libffi-dev libgdbm3 libgdbm-dev cmake libxml2 \
    libxml2-dev flex pkg-config libglib2.0-dev \
    libcairo-dev libpango1.0-dev libgdk-pixbuf2.0-dev \
    libpangocairo-1.0

# Install rbenv from https://github.com/rbenv/rbenv
git clone https://github.com/rbenv/rbenv.git ~/.rbenv

# Set path to shim layers in .bashrc
echo 'export PATH="$HOME/.rbenv/bin:$PATH"' >> .bashrc

~/.rbenv/bin/rbenv init

# Set .rbenv environment variables in .bashrc
echo 'eval "$(rbenv init -)"' >> .bashrc

# Restart your shell (e.g. open a new terminal window). Note that
# you do not need to use the `-l` option, since the modifications
# were made to .bashrc rather than .bash_profile. If successful,
# `type rbenv` should print 'rbenv is a function' followed by code.

# Install `ruby_build` plugin from https://github.com/rbenv/ruby-build

git clone https://github.com/rbenv/ruby-build.git ~/.rbenv/plugins/ruby-build

# Install Ruby 2.3.3
# This takes in excess of 20 min. to build!
# https://github.com/rbenv/ruby-build/issues/1054#issuecomment-276934761
# suggests:
# "You can speed up Ruby installs by avoiding generating ri/RDoc
# documentation for them:
# RUBY_CONFIGURE_OPTS=--disable-install-doc rbenv install 2.3.3
# We have not tried this.

rbenv install 2.3.3

# Configure rbenv globally to always use Ruby 2.3.3.
echo "2.3.3" > ~/.rbenv/version

# Finally, install toolchain components.
# asciidoctor-mathematical also takes in excess of 20 min. to build!
# The same RUBY_CONFIGURE_OPTS advice above may apply here as well.

gem install asciidoctor coderay json-schema
gem install --pre asciidoctor-pdf
MATHEMATICAL_SKIP_STRDUP=1 gem install asciidoctor-mathematical
----


[[depends-ubuntu-rvm]]
===== Ubuntu/Windows 10 Using RVM

Here are (sparser) instructions for using rvm to setup version 2.3.x:

----
gpg --keyserver hkp://keys.gnupg.net --recv-keys 409B6B1796C275462A1703113804BB82D39DC0E3
\curl -sSL https://get.rvm.io | bash -s stable --ruby
source ~/.rvm/scripts/rvm
rvm install ruby-2.3
rvm use ruby-2.3
----

NOTE: Windows 10 Bash will need to be launched with the "-l" option
appended, so that it runs a login shell; otherwise RVM won't function
correctly on future launches.


[[depends-ubuntu-sys]]
===== Ubuntu 16.04 using system Ruby

The Ubuntu 16.04.1 default Ruby install (version 2.3.1) seems to be
up-to-date enough to run all the required gems, but also needs the
`ruby-dev` package installed through the package manager.

In addition, the library
`/var/lib/gems/2.3.0/gems/mathematical-1.6.7/ext/mathematical/lib/liblasem.so`
has to be copied or linked into a directory where the loader can find it.
This requirement appears to be due to a problem with the
asciidoctor-mathematical build process.


[[depends-mingw]]
==== MinGW

MinGW can be obtained here: http://www.mingw.org/

Once the installer has run its initial setup, following the
http://www.mingw.org/wiki/Getting_Started[instructions on the website], you
should install the `mingw-developer-tools`, `mingw-base` and `msys-base`
packages.
The `msys-base` package allows you to use a bash terminal from windows with
whatever is normally in your path on Windows, as well as the unix tools
installed by MinGW.

In the native Windows environment, you should also install the following
native packages:

  * Python 3.x (https://www.python.org/downloads/)
  * Ruby 2.x (https://rubyinstaller.org/)
  * Git command-line client (https://git-scm.com/download)

Once this is setup, and the necessary <<depends-gems,Ruby Gems>> are
installed, launch the `msys` bash shell, and navigate to the spec Makefile.
From there, you'll need to set `PYTHON=` to the location of your python
executable for version 3.x before your make command - but otherwise
everything other than pdf builds should just work.

NOTE: Building the PDF spec via this path has not yet been tested but *may*
be possible - liblasem is the main issue and it looks like there is now a
mingw32 build of it available.


[[depends-cygwin]]
==== Cygwin

When installing Cygwin, you should install the following packages via
`setup`:

----
// "curl" is only used to download fonts, can be done in another way
autoconf
bison
cmake
curl
flex
gcc-core
gcc-g++
ghostscript
git
libbz2-devel
libcairo-devel
libcairo2
libffi-devel
libgdk_pixbuf2.0-devel
libiconv
libiconv-devel
liblasem0.4-devel
libpango1.0-devel
libpango1.0_0
libxml2
libxml2-devel
make
python3
ruby
ruby-devel
----

NOTE: Native versions of some of these packages are usable, but care should
be taken for incompatibilities with various parts of cygwin - e.g. paths.
Ruby in particular is unable to resolve Windows paths correctly via the
native version.
Python and Git for Windows can be used, though for Python you'll need to set
the path to it via the PYTHON environment variable, before calling make.

When it comes to installing the mathematical ruby gem, there are two things
that will require tweaking to get it working.
Firstly, instead of:

----
MATHEMATICAL_SKIP_STRDUP=1 gem install asciidoctor-mathematical
----

You should use

----
MATHEMATICAL_USE_SYSTEM_LASEM=1 gem install asciidoctor-mathematical
----

The latter causes it to use the lasem package already installed, rather than
trying to build a fresh one.

The mathematical gem also looks for "liblasem" rather than "liblasem0.4" as
installed by the lasem0.4-devel package, so it is necessary to add a symlink
to your /lib directory using:

----
ln -s /lib/liblasem-0.4.dll.a /lib/liblasem.dll.a
----

<<Ruby Gems>> are not installed to a location that is in your path normally.
Gems are installed to `~/bin/` - you should add this to your path before
calling make:

    export PATH=~/bin:$PATH

Finally, you'll need to manually install fonts for lasem via the following
commands:

----
mkdir /usr/share/fonts/truetype cd /usr/share/fonts/truetype
curl -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/cmex10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/cmmi10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/cmr10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/cmsy10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/esint10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/eufm10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/msam10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/msbm10.ttf
----


[[depends-osx]]
=== Mac OS X

Mac OS X should work in the same way as for ubuntu by using the Homebrew
package manager, with the exception that you can simply install the ruby
package via `brew` rather than using a ruby-specific version manager.

You'll likely also need to install additional fonts for the PDF build via
mathematical, which you can do with:

----
cd ~/Library/Fonts
curl -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/cmex10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/cmmi10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/cmr10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/cmsy10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/esint10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/eufm10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/msam10.ttf \
     -LO http://mirrors.ctan.org/fonts/cm/ps-type1/bakoma/ttf/msbm10.ttf
----

Then install the required <<depends-gems,Ruby Gems>>.


[[depends-linux]]
=== Linux (Debian, Ubuntu, etc.)

The instructions for the <<depends-ubuntu,Ubuntu / Windows 10>> installation
are generally applicable to native Linux environments using Debian packages,
such as Debian and Ubuntu, although the exact list of packages to install
may differ.
Other distributions using different package managers, such as RPM (Fedora)
and Yum (SuSE) will have different requirements.

Using `rbenv` or `rvm` is neccessary, since the system Ruby packages are
often well out of date.

Once the environment manager, Ruby, and `ruby_build` have been installed,
install the required <<depends-gems,Ruby Gems>>.


[[depends-gems]]
=== Ruby Gems

The following ruby gems can be installed directly via the `gem install`
command, once the platform is set up:

----
gem install rake asciidoctor coderay json-schema

# Required only for pdf builds
MATHEMATICAL_SKIP_STRDUP=1 gem install asciidoctor-mathematical
gem install --pre asciidoctor-pdf
----

[[ruby-enum-downgrade]]
==== Ruby Gem Versioning Errors

*ruby-enum*

As of 2017-03-06, there appears to be a problem with the ruby-enum version
0.7.1 gem which breaks the PDF build. Make sure you are using ruby-enum
0.7.0, as follows:

    gem uninstall ruby-enum
    gem install -v 0.7.0 ruby-enum

Hopefully this will soon be fixed. See
https://github.com/gjtorikian/mathematical/issues/69 for a report of this
problem.


*prawn*

As of 2017-03-20, there are incompatibilities between asciidoctor-pdf and
certain versions of prawn and prawn-templates affecting the PDF build. Make
sure to update to prawn 2.2.1 and prawn-templates 0.0.5. See

https://github.com/KhronosGroup/Vulkan-Docs/issues/476


[[history]]
== Revision History

  * 2017-03-20 - Add description of prawn versioning problem and how to fix
    it.
  * 2017-03-06 - Add description of ruby-enum versioning problem and how to
    fix it.
  * 2017-02-13 - Move some comments here from ../../../README.md. Tweak
    asciidoctor markup to more clearly delineate shell command blocks.
  * 2017-02-10 - Add more Ruby installation guidelines and reflow the
    document in accordance with the style guide.
  * 2017-01-31 - Add rbenv instructions and update the README elsewhere.
  * 2017-01-16 - Modified dependencies for Asciidoctor
  * 2017-01-06 - Replace MathJax with KaTeX.
  * 2016-08-25 - Update for the single-branch model.
  * 2016-07-10 - Update for current state of spec and ref page generation.
  * 2015-11-11 - Add new can: etc.
    macros and DBLATEXPREFIX variable.
  * 2015-09-21 - Convert document to asciidoc and rename to README.md in the
    hope the gitlab browser will render it in some fashion.
  * 2015-09-21 - Add descriptions of LaTeX and MathJax math support for all
    output formats.
  * 2015-09-02 - Added Cygwin package info.
  * 2015-09-02 - Initial version documenting macros, required toolchain
    components and versions, etc.