You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

76 lines
4.4 KiB

Deterministic OSX Dmg Notes.
Working OSX DMG's are created in Linux by combining a recent clang,
the Apple's binutils (ld, ar, etc), and DMG authoring tools.
Apple uses clang extensively for development and has upstreamed the necessary
functionality so that a vanilla clang can take advantage. It supports the use
of -F, -target, -mmacosx-version-min, and --sysroot, which are all necessary
when building for OSX. A pre-compiled version of 3.2 is used because it was not
available in the Precise repositories at the time this work was started. In the
future, it can be switched to use system packages instead.
Apple's version of binutils (called cctools) contains lots of functionality
missing in the FSF's binutils. In addition to extra linker options for
frameworks and sysroots, several other tools are needed as well such as
install_name_tool, lipo, and nmedit. These do not build under linux, so they
have been patched to do so. The work here was used as a starting point:
In order to build a working toolchain, the following source packages are needed
from Apple: cctools, dyld, and ld64.
Beware. This part is ugly. Very very very ugly. In the future, this should be
broken out into a new repository and cleaned up. Additionally, the binaries
only work when built as x86 and not x86_64. This is an especially nasty
limitation because it must be linked with the toolchain's, meaning
that the entire toolchain must be x86. Gitian x86_64 should not be used until
this has been fixed, because it would mean that several native dependencies
(openssl, libuuid, etc) would need to be built as x86 first.
These tools inject timestamps by default, which produce non-deterministic
binaries. The ZERO_AR_DATE environment variable is used to disable that.
This version of cctools has been patched to use the current version of clang's
headers and and its rather than those from llvmgcc, as it was
originally done in toolchain4.
To complicate things further, all builds must target an Apple SDK. These SDKs
are free to download, but not redistributable.
To obtain it, register for a developer account, then download xcode4630916281a.dmg:
This file is several gigabytes in size, but only a single directory inside is
Unfortunately, the usual linux tools (7zip, hpmount, loopback mount) are incapable of opening this file.
To create a tarball suitable for gitian input, mount the dmg in OSX, then create it with:
$ tar -C /Volumes/Xcode/ -czf MacOSX10.7.sdk.tar.gz MacOSX10.7.sdk
The gitian descriptors build 2 sets of files: Linux tools, then Apple binaries
which are created using these tools. The build process has been designed to
avoid including the SDK's files in Gitian's outputs. All interim tarballs are
fully deterministic and may be freely redistributed.
genisoimage is used to create the initial DMG. It is not deterministic as-is,
so it has been patched. A system genisoimage will work fine, but it will not
be deterministic because the file-order will change between invocations.
The patch can be seen here:
No effort was made to fix this cleanly, so it likely leaks memory badly. But
it's only used for a single invocation, so that's no real concern.
genisoimage cannot compress DMGs, so afterwards, the 'dmg' tool from the
libdmg-hfsplus project is used to compress it. There are several bugs in this
tool and its maintainer has seemingly abandoned the project. It has been forked
and is available (with fixes) here: .
The 'dmg' tool has the ability to create DMG's from scratch as well, but this
functionality is broken. Only the compression feature is currently used.
Ideally, the creation could be fixed and genisoimage would no longer be necessary.
Background images and other features can be added to DMG files by inserting a
.DS_Store before creation. The easiest way to create this file is to build a
DMG without one, move it to a device running OSX, customize the layout, then
grab the .DS_Store file for later use. That is the approach taken here.