This is where it all begins.
A kernel bootstrapping option is used by default at the beginning. The tiny builder-hex0-x86-stage1 binary seed boots, builds and runs the builder-hex0-x86-stage2 kernel. The builder-hex0-x86-stage1 and builder-hex0-x86-stage2 kernels are written in hex0. The stage2 kernel has its own built-in shell, the hex0 compiler and the src tool to load files into its memory file system. In this case the first step is to build the hex0-seed and kaem-optional-seed binaries from hex0 source.
Note that all early shells and compilers before mes are part of stage0-posix.
builder-hex0-x86-stage2 runs as the kernel for building the stage0-posix compilers and then mes and tcc. Then the Fiwix <https://github.com/mikaku/Fiwix> kernel is built and launched. Then Fiwix runs the build until Linux takes over.
If chroot or bwrap is specified or if a pre-existing kernel is provided then we start with the two raw binary seeds hex0-seed and kaem-optional-seed. We use those seeds to rebuild themselves.
hex0 is fairly trivial to implement and for each pair of hexadecimals characters it outputs a byte. We have also added two types of line comments (# and ;) to create a well commented lines like:
# :loop_options [_start + 0x6F]
39D3 ; cmp_ebx,edx # Check if we are done
74 14 ; je !loop_options_done # We are done
83EB 02 ; sub_ebx, !2 # --optionsIn the first steps we use initial hex0 binary seed to rebuild kaem-optional and hex0 from their source.
hex0 code is somewhat tedious to read and write as it is basically a well documented machine code. We have to manually calculate all jumps in the code.
hex0 can be approximated with: sed 's/[;#].*$//g' $input_file | xxd -r -p > $output_file
By default (when kernel bootstrap is enabled), the builder-hex0-x86-stage1 boot loader/compiler boots from a hard drive and loads hex0 source code from disk, compiles, and runs the builder-hex0-x86-stage2 kernel. builder-hex0-x86-stage1 is written in hex0 and can be compiled with any one of hex0-seed, sed, or the tiny builder-hex0-mini binary.
When kernel bootstrap is enabled, the builder-hex0-x86-stage2 kernel loads an enormous shell script which embeds files (loaded with the src command) and the initial commands to build hex0-seed, kaem-optional-seed, and the command which launches stage0-posix using kaem-optional-seed and the stage0-posix launch script kaem.x86.
kaem-optional is a trivial shell that can read list of commands together with their command line arguments from a file and executes them. It also supports line comments but has no other features.
This is the last program that has to be written in hex0 language. hex1 is a simple extension of hex0 and adds a single character labels and allows calculating 32-bit offsets from current position in the code to the label. hex1 code might look like:
:a #:loop_options
39D3 ; cmp_ebx,edx # Check if we are done
0F84 %b ; je %loop_options_done # We are done
83EB 02 ; sub_rbx, !2 # --optionshex2 is our final hex language that adds support for labels of arbitrary length. It also allows accessing them via 8, 16, 32-bit relative addresses (!, @, %) and via 16-bit or 32-bit ($, &) absolute addresses:
:loop_options
39D3 ; cmp_ebx,edx # Check if we are done
74 !loop_options_done ; je8 !loop_options_done # We are done
83EB 02 ; sub_ebx, !2 # --optionscatm allows concatenating files with catm output_file input1 input2 ... inputN. This allows us to distribute common code in separate files. We will first use it to append ELF header to .hex2 files. Before this step the ELF header had to be included in the source file itself.
The M0 assembly language is the simplest assembly language you can create that enables the creation of more complicated programs. It includes only a single keyword: DEFINE and leverages the language properties of hex2 along with extending the behavior to populate immediate values of various sizes and formats.
Thus M0 code looks like:
DEFINE cmp_ebx,edx 39D3
DEFINE je 0F84
DEFINE sub_ebx, 81EB
:loop_options
cmp_ebx,edx # Check if we are done
je %loop_options_done # We are done
sub_ebx, %2 # --optionsThe cc_x86 implements a subset of the C language designed in M0 assembly. It is a somewhat limited subset of C but complete enough to make it easy to write a more usable C compiler written in the C subset that cc_x86 supports.
At this stage we start using M2libc as our C library. In fact, M2libc ships two versions of C library. There is a single-file library that contains just enough to build M2-Planet and there is a full version that is rather well-featured.
This is the only C program that we build with cc_x86. M2-Planet supports a larger subset of C than cc_x86 and we are somewhat closer to C89 (it does not implement all C89 features but on the other hand it does have some C99 features). M2-Planet also includes a very basic preprocessor, so we can use stuff like #define, #ifdef.
M2-Planet is also capable of using full M2libc C library that has more features and optimizations compared to bootstrap version of M2libc.
M2-Planet supports generating code for various architectures including x86, amd64, armv7, aarch64, riscv32 and riscv64. Up until this point bootstrap has been very architecture specific. From now on we still have platform specific bits of code but they are usually handled as conditionals in the same application rather than having completely different applications.
Now we build blood-elf used to generate debug info, C version of hex2 (also called hex2) and C version of M0 called M1. These are more capable than their platform specific hex counterparts and are fully cross-platform. Thus we can now have the whole toolchain written in C.
Then we rebuild mescc-tools again, so all our tools are using new toolchain written in C.
Finally, we build kaem which is a more capable version of kaem-optional and adds support for variables, environmental variables, conditionals and aliases. It also has various built-ins such as cd and echo.
M2-Mesoplanet is a preprocessor that is more capable than M2-Planet and supports #include statements. It can also launch compiler, assembler and linker with the correct arguments, so we don't need to invoke them
manually.
At the moment it is only used to build mescc-tools-extra.
We rebuild M2-Planet with M2-Planet.
From here, we can move on from the lowest level stuff.
mescc-tools-extra contains some additional programs, namely filesystem
utilities cp and chown. This allows us to have one unified
directory for our binaries. Furthermore, we also build sha256sum, a
checksumming tool, that we use to ensure reproducibility and authenticity
of generated binaries. We also build initial untar, ungz, unxz
and unbz2 utilities to deal with compressed archives, as well as
replace, a trivial search-and-replace program.
stage0-posix executes a file after.kaem, which creates a kaem script to
continue the bootstrap. This is responsible for cleaning up the mess in
/x86/bin and moving it to the permanent /usr/bin, and setting a few
environment variables.
script-generator is a program that translates live-bootstrap's
domain-specific manifest language into shell scripts that can be run to complete
the bootstrap. The translator is implemented in M2-Planet.
The language is fairly simple; each line has the format
<directive>: <arguments> <predicate>. A predicate only runs the line if a
particular condition is true.
The following directives are supported:
build, builds a particular package defined insteps/.improve, runs a script making a distinct and logical improvement to the live bootstrap system.define, defines a variable evaluated from other constants/variables.jump, moves into a new rootfs/kernel using a custom script.uninstall, removes a previously built package for file.
checksum-transcriber is a small program that converts live-bootstrap's
source specification for packages into a SHA256SUM file that can be used to
checksum source tarballs.
simple-patch is a rudimentary patching program. It works by matching for a
text block given to it, and replacing it with another text block. This is
sufficient for the early patching required before we have full proper GNU patch.
GNU mes is a scheme interpreter. It runs the sister project mescc,
which is a C compiler written in scheme, which links against the Mes C
Library. All 3 are included in this same repository. There are two stages
to this part:
- Compiling an initial mes using
M2-Planet. Note that this is only the Mes interpreter, not the libc or anything else. - We then use this to recompile the Mes interpreter as well as building the libc. This second interpreter is faster and less buggy.
The mescc component depends on the nyacc parsing library, version 1.00.2.
We use a modified version, 1.00.2-lb1, which incorporates Timothy Sample's
changes required to bootstrap mescc without relying on pregenerated files.
From this point until musl, we are capable of making non-standard and strange
libraries. All libraries are in /usr/lib/mes, and includes are in
/usr/include/mes, as they are incompatible with musl.
tinycc is a minimal C compiler that aims to be small and fast. It
complies with all C89 and most of C99 standards.
First, we compile janneke’s fork of tcc 0.9.26 using mescc,
containing 27 patches to make it operate well in the bootstrap
environment and make it compilable using mescc. This is a
non-trivial process and as seen within tcc. kaem has many different parts
within it: a. tcc 0.9.26 is first compiled using mescc. b. The mes
libc is recompiled using tcc (mescc has a non-standard .a
format), including some additions for later programs. c. tcc 0.9.26 is
recompiled 3 times to add new features, namely long long and
float. Each time, the libc is also recompiled.
Now, we compile upstream tcc 0.9.27, the latest release of tinycc, using the final version of tcc 0.9.26.
From this point onwards, until further notice, all programs are compiled using tinycc 0.9.27.
Note that now we begin to delve into the realm of old GNU software, using older versions compilable by tinycc. Prior to this point, all tools have been adapted significantly for the bootstrap; now, we will be using old tooling instead.
If the kernel bootstrap option is enabled then the Fiwix kernel is built next. This is a Linux 2.0 clone which is much simpler to understand and build than Linux. This version of Fiwix is an intermediate release on top of 1.5.0 that contains many modifications and enhancements to support live-boostrap.
If the kernel bootstrap option is enabled then lwext4 <https://github.com/gkostka/lwext4>
is built next. This is a library for creating ext2/3/4 file systems from user land.
This is combined with a program called make_fiwix_initrd.c which creates
and populates an ext2 files system which Fiwix uses for an initial ram drive (initrd).
This file system contains all of the files necessary to build Linux.
If the kernel bootstrap option is enabled then a C program kexec-fiwix is compiled and run which places the Fiwix ram drive in memory and launches the Fiwix kernel.
GNU make is now built so we have a more robust building system.
make allows us to do things like define rules for files rather than
writing complex kaem scripts.
patch is a very useful tool at this stage, allowing us to make
significantly more complex edits, including just changes to lines.
gzip is the most common compression format used for software source
code. It is more capable than ungz from stage0-posix and also supports
compression.
We build GNU Tar 1.12, the last version compilable with mes libc.
You are most likely aware of GNU sed, a line editor.
bzip2 is a compression format that compresses more than gzip. It
is preferred where we can use it, and makes source code sizes smaller.
GNU Coreutils is a collection of widely used utilities such as cat,
chmod, chown, cp, install, ln, ls, mkdir,
mknod, mv, rm, rmdir, tee, test, true, and
many others.
A few of the utilities cannot be easily compiled with Mes C library, so we skip them.
The cp in this stage replaces the mescc-tools-extra cp.
The Berkeley Yacc parser generator, a public-domain implementation of the
yacc utility. Differently from the analogous bison utility from the
GNU project, it can be compiled with a simple Makefile.
Some code is backported from an earlier version of byacc, 20140101, because of an incompatibility of newer versions with meslibc.
GNU bash is the most well known shell and the most complex piece of
software so far. However, it comes with a number of great benefits over
kaem, including proper POSIX sh support, globbing, etc.
Bash ships with a bison pre-generated file here which we delete. Unfortunately, we have not bootstrapped bison but fortunately for us, Berkeley Yacc is able to cope here.
This is a simple script that sets up the /external/repo directory to hold
binary tarballs of artifacts built in each step. It also creates base.tar.bz2,
a tarball containing every artifact built before setup_repo, which have no
individual repository tarballs corresponding to them.
From this point on, every package is built and installed into a temporary directory, packaged from there into a repository tarball, and then installed onto the live system from the newly built repository tarball.
This is a simple script that makes some small updates to the env file that were not possible when using kaem.
Sets up symlinks from folders outside the /usr hierarchy to the corresponding
/usr ones ("merged-usr" file system layout).
Sets up important device nodes under /dev. These nodes are temporary, as once
the Linux kernel is started, devtmpfs is used to maintain /dev.
In interactive mode only, sets up an interactive Bash console, accessible by pressing Ctrl+Alt+F2. Because the early Bash doesn't support true interactive operation, we emulate it using a REPL. A side effect of this is that after every command entered, one must press Enter followed by Ctrl+D, rather than just Enter as with a real interactive shell.
We recompile tcc with some patches needed to build musl.
musl is a C standard library that is lightweight, fast, simple,
free, and strives to be correct in the sense of standards-conformance
and safety. musl is used by some distributions of GNU/Linux as their
C library. Our previous Mes C library was incomplete which prevented us
from building many newer or more complex programs.
tcc has slight problems when building and linking musl, so we
apply a few patches.
We do not use any of /usr/lib/mes or /usr/include/mes any longer, rather
using /usr/lib and /usr/include like normal.
We recompile tcc against musl. This is a two stage process. First we
build tcc-0.9.27 using tcc-0.9.26 that itself links to Mes C library but produces
binaries linked to musl. Then we recompile newly produced tcc with
itself. Interestingly, tcc-0.9.27 linked against musl is self hosting.
We now rebuild musl with the just built tcc-musl, which fixes a
number of bugs, particularly regarding floats, in the first musl.
Now that we have a ‘fixed’ musl, we now recompile tcc as tcc
uses floats extensively.
sed is rebuilt against musl.
bzip2 is rebuilt unpatched with the new tcc and musl fixing issues
with reading files from stdin that existed in the previous build.
m4 is the first piece of software we need in the autotools suite,
flex 2.6.4 and bison. It allows macros to be defined and files to be
generated from those macros.
lex from the Heirloom project. The Heirloom project is a collection
of standard UNIX utilities derived from code by Caldera and Sun.
Differently from the analogous utilities from the GNU project, they can
be compiled with a simple Makefile.
Because issues with the Heirloom version of yacc compiled against musl,
we continue using Berkeley Yacc together with Heirloom lex for the next
few steps.
flex is a tool for generating lexers or scanners: programs that
recognize lexical patterns.
Unfortunately flex also depends on itself for compiling its own
scanner, so first flex 2.5.11 is compiled, with its scanner definition
manually modified so that it can be processed by lex from the Heirloom
project (the required modifications are mostly syntactical, plus a few
workarounds to avoid some flex advanced features).
We recompile unpatched GNU flex using older flex 2.5.11. This is
again a two stage process, first compiling flex using scan.c (from
scan.l) created by old flex, then recompile scan.c using the new
version of flex to remove any buggy artifacts from the old flex.
GNU bison is a parser generator. With m4 and flex we can now
bootstrap it following https://gitlab.com/giomasce/bison-bootstrap. It’s
a 3 stage process:
- Build bison using a handwritten grammar parser in C.
- Use bison from previous stage on a simplified bison grammar file.
- Build bison using original grammar file.
Finally we have a fully functional bison executable.
GNU grep is a pattern matching utility. Is is not immediately needed
but will be useful later for autotools.
diffutils is useful for comparing two files. It is not immediately
needed but is required later for autotools.
coreutils is rebuilt against musl. Additional utilities are built
including comm, expr, dd, sort, sync, uname and
uniq. This fixes a variety of issues with existing coreutils.
We build date, mktemp and sha256sum from coreutils 6.10 which are
either missing or don't build correctly in 5.0. Other utils are not built at
this stage.
gawk is the GNU implementation of awk, yet another pattern
matching and data extraction utility. It is also required for autotools.
Perl is a general purpose programming language that is especially suitable for text processing. It is essential for autotools build system because automake and some other tools are written in Perl.
Perl itself is written in C but ships with some pre-generated files that
need perl for processing, namely embed.h and keywords.h. To
bootstrap Perl we will start with the oldest Perl 5 version which has
the fewest number of pregenerated files. We reimplement two remaining
perl scripts in awk and use our custom makefile instead of Perl’s
pre-generated Configure script.
At this first step we build miniperl which is perl without
support for loading modules.
We now use perl from the previous stage to recreate pre-generated
files that are shipped in perl 5.003. But for now we still need to use
handwritten makefile instead of ./Configure script.
Yet another version of perl; the last version buildable with 5.003.
More perl! This is the last version buildable with 5.004. It also
introduces the new pregenerated files regnodes.h and
byterun.{h,c}.
Even more perl. 5.6.2 is the last version buildable with 5.005.
GNU Autoconf is a tool for producing configure scripts for building, installing and
packaging software on computer systems where a Bourne shell is available.
At this stage we still do not have a working autotools system, so we manually install
autoconf script and replace a few placeholder variables with sed.
Autoconf 2.52 is the newest version of autoconf that does not need perl, and hence
a bit easier to install.
This is not a full featured autoconf install, it is missing other programs such as autoheader
but is sufficient to build autoconf 2.53.
GNU Automake is a tool for automatically generating Makefile.in files. It is another major
part of GNU Autotools build system and consists of aclocal and automake scripts.
We bootstrap it using a 2 stage process:
- Use
sedto replace a few placeholder variables inaclocal.inscript. Then we manually installaclocalscript and its dependencies. - Patch
configure.into createautomakefile but skipMakefile.inprocessing. Again we manually installautomakescript and its dependencies.
We now start bootstrapping newer versions of autoconf. Version 2.53 now uses perl.
In order to build it with autoconf-2.52 we have to patch it a bit.
Automake 1.7 and Autoconf 2.54 depend on each other, so we patch out two offending
autoconf macros to make it build with autoconf-2.53.
Never version of autoconf.
Even newer autoconf. This is the last version of autoconf that is buildable
with automake-1.7.
Newer automake. This is the latest automake that is buildable with autoconf-2.55.
Newer autoconf. This time we were able to skip version 2.56.
Again, we managed to skip one version.
We need newer automake to proceed to newer autoconf versions. This is the latest
automake version from 1.8 release series.
help2man automatically generates manpages from programs --help and --version
outputs. This is not strictly required for bootstrapping but will help us to avoid patching
build process to skip generation of manpages. This is the newest version of help2man that
does not require Perl 5.8.
Yet another version of autoconf.
Latest GNU Automake from 1.9 series. Slightly annoyingly depends on itself but it is easy to patch to make it buildable with 1.8.5.
GNU Automake from 1.10 series. aclocal is slightly patched to work
with our perl.
Slightly newer version of GNU Autoconf. At this stage Autoconf is mostly
backwards compatible but newer versions need newer automake.
GNU Automake from 1.11 series. This is not the latest point release as newer ones
need Autoconf 2.68. Newer major version of automake also depends on a newer bash.
This is a much newer version of GNU Autoconf.
GNU Libtool is the final part of GNU Autotools. It is a script used to hide away differences when compiling shared libraries on different platforms.
GNU Automake from 1.15 series. This is the last version that runs on Perl 5.6.
The GNU Binary Utilities, or binutils, are a set of programming tools for creating and managing binary programs, object files, libraries, profile data, and assembly source code.
In particular we can now use full featured ar instead of tcc -ar,
the GNU linker ld, which allows us building shared libraries,
and the GNU assembler as.
We rebuild musl for the third time. This time we can use GNU as to build assembly source files, so those assembly files that tcc failed to compile no longer have to be patched.
We rebuild tcc against new musl.
The GNU Compiler Collection (GCC) is an optimizing compiler produced by the GNU Project. GCC is a key component of the GNU toolchain and the standard compiler for most projects related to GNU and the Linux kernel.
Only the C frontend is built at this stage.
At this stage we are not yet able to regenerate top-level Makefile.in
which needs GNU Autogen and hence Guile. Luckily, building GCC without
top-level Makefile is fairly easy.
GNU Find Utilities can be used to search for files. We are mainly interested
in find and xargs that are often used in scripts.
GCC can build the latest as of the time of writing musl version.
We also don't need any of the TCC patches that we used before. To accomodate Fiwix, there are patches to avoid syscalls set_thread_area and clone.
This gets some headers out of the Linux kernel that are required to use the
kernel ABI, needed for util-linux.
The version of the Linux kernel used comes from the Open Enterprise Linux Association, who maintain this version as a continuation of the now ended 4.14 LTS release, to the same maintenance standards as the LTS. Because this isn't directly available as an efficiently compressed tarball, we start with the final LTS release, version 4.14.336, and apply the differences using a patch file.
Rebuild GCC with GCC and also against the latest musl.
util-linux contains a number of general system administration utilities.
This gives us access to a much less crippled version of mount and mknod.
The latest version is not used because of autotools/GCC incompatibilities.
dhcpcd is a DHCP client daemon, which we later use to obtain an IP address.
Only wired interfaces are supported at the moment.
kbd contains loadkeys which is required for building the Linux kernel.
The 2.x series is not used because it requires particular features of autotools
that we do not have available.
GNU make is now rebuilt properly using the build system and GCC, which means that
it does not randomly segfault while building the Linux kernel.
ed is a very basic line editor. This is the last version that is not distributed
in .tar.lz format. ed is used by bc build scripts.
bc is a console based calculator that is sometime used in scripts. We need bc
to rebuild some Linux kernel headers.
If the kernel bootstrap option is enabled then a C program kexec-linux is compiled.
This can be used to launch a Linux kernel from Fiwix (when not using --kernel).
kexec is a utility for the Linux kernel that allows the re-execution of the
Linux kernel without a manual restart from within a running system. It is a
kind of soft-restart. It is only built for non-chroot mode, as we only use it
in non-chroot mode. It is used to boot the Linux kernel that will be built next
from the current Linux kernel (when using --kernel).
A script to remove source tarballs no longer needed for subsequent build steps from distfiles. This frees space for the Linux kernel build.
More space freeing in preparation for the kernel build. This time, we clear out any
artifacts left over from previous builds that weren't packaged in a proper repository
tarball or base.tar.bz2
A lot going on here. This is the first (and currently only) time the Linux kernel
is built. Firstly, Linux kernel version 4.14.y is used because newer versions
require much more stringent requirements on the make and GCC versions. This is also
modern enough for most hardware and to cause few problems with software built
afterwards. Secondly, since 4.14.y is no longer supported by kernel.org, we use the
last available kernel.org tarball, 4.14.336, and patch it up to 4.14.341-openela,
maintained by the Open Enterprise Linux Association as an ELTS version.
Pregenerated files in the kernel have file names appended with _shipped
so we use a find command to remove those, which are automatically regenerated.
The kernel config was originally taken from Void Linux, and was then modified
for the requirements of live-bootstrap, including compiler features, drivers,
and removing modules. Modules are unused. They are difficult to transfer to
subsequent systems, and we do not have modprobe.
The linux-libre scripts are no longer used to deblob the kernel, due to undesirable modifications they make beyond just deblobbing. Instead, the remaining 4 drivers that ship binary blobs in line with source code are stripped using a patch - neither of these drivers are relevant to bootstrapping.
The kernel is built in 2 stages:
- We build
vmlinuxand all of its dependencies. - After clearing away any unnecessary intermediate files, we build the final
bzImagekernel. This is necessary because the whole build wouldn't fit in Fiwix's initrd image all at once.
We then kexec to use the new Linux kernel, using kexec-tools for a Linux
kernel and kexec-linux for Fiwix.
In kernel-bootstrap mode, we have been working off an initramfs for some
things up until now. At this point we are now capable of moving to it entirely,
so we do so.
In kernel-bootstrap mode, up until this point, we had no multiprocessor
support, and very limited RAM, so all builds used only one thread.
At this point, we allow the full selected thread count to take effect,
speeding up subsequent builds thanks to parallelization.
Sets up the file system as per the Filesystem Hierarchy Standard (FHS), creating directories and mounting pseudo-filesystems as necessary.
In interactive mode only, sets up an interactive Bash console, accessible by pressing Ctrl+Alt+F2, again. This is still the early Bash, requiring Ctrl+D.
If enabled in bootstrap.cfg, creates and activates a swap file under the
name /swapfile.
At this point, it is guaranteed that we are running on Linux with thread support, so we rebuild musl with thread support.
curl is used to download files using various protocols including HTTP and HTTPS.
However, this first build does not support encrypted HTTPS yet. curl requires
Linux and musl with thread support, which are now available.
Once curl is built, we use dhcpcd to set up networking for downloading subsequent source packages.
This new version of bash compiles without any patches, provides new features,
and is built with GNU readline support so it can be used as a fully-featured
interactive shell. autoconf-2.69 is used to regenerate the configure
script and bison is used to recreate some included generated files.
Now that we have a proper interactive shell available, open another interactive console (only in interactive mode), this time accessible using Ctrl+Shift+F3, since Ctrl+Shift+F2 is already occupied by our previous console, running the old Bash.
XZ Utils is a set of free software command-line lossless data compressors,
including lzma and xz. In most cases, xz achieves higher compression rates
than alternatives like gzip and bzip2.
file is a utility that is used to get information about files based upon their magic.
A modern version of libtool with better compatibility with newer versions of GNU Autotools.
Newer tar has better support for decompressing .tar.bz2 and .tar.xz archives. It also deals better with modern tar archives with extra metadata.
We build the latest available coreutils 9.4 which adds needed options to make
results of build metadata reproducible. For example, timestamps are changed with
touch --no-dereference.
pkg-config is a helper tool that helps to insert compile and link time flags.
A newer version of make built using autotools is much more reliable and is compiled using a modern C compiler and C library. This removes a couple of segfaults encountered later in the process and allows more modern make features to be used. We do not go for the latest because of the use of automake 1.16 which we do not have yet.
GNU Multiple Precision Arithmetic Library (GMP) is a free library for arbitrary-precision arithmetic, operating on signed integers, rational numbers, and floating-point numbers.
GMP is required by newer versions of GCC and Guile.
The GNU Autoconf Archive is a collection of Autoconf macros that are used by various projects and in particular GNU MPFR.
The GNU Multiple Precision Floating-Point Reliable Library (GNU MPFR) is a library for arbitrary-precision binary floating-point computation with correct rounding, based on GNU Multi-Precision Library.
GNU MPC is a library for multiprecision complex arithmetic with exact rounding based on GNU MPFR.
An older version of flex is required for bison 2.3. We cannot use 2.5.11 that was compiled much earlier, as it does not produce reproducible output when building bison 2.3.
This is an older version of bison required for the bison files in perl 5.10.1. We backwards-bootstrap this from 3.4.1, using 3.4.1 to compile the bison files in 2.3. This parser works sufficiently well for perl 5.10.1.
Bison 3.4.1 is buggy and segfaults when perl 5.32.1 is built. This is probably because it was built with a hand-written makefile. We do not build the latest bison because perl 5.32.1 requires bison <= 3.4.2.
Perl 5.10.1 is an intermediate version used before Perl 5.32. We require this version as it adds a couple of modules into lib/ required to regenerate files in Perl 5.32. We still use the Makefile instead of the metaconfig strategy, as metaconfig history becomes poor more than a few years back.
dist is perl's package used for generating Perl's Configure (which is written in Perl itself). We 'compile' (aka generate) metaconfig and manifake only from dist. We do not use dist's build system because it itself uses dist.
We finally compile a full version of Perl using Configure. This includes all base extensions required and is the latest version of Perl. We are now basically able to run any Perl application we want.
libarchive is a C library used to read and write archives.
OpenSSL is a C library for secure communications/cryptography.
We do not use the latest 3.3.0 release because it causes lockups in curl.
Install TLS root certificates from nss. This will allows us to use HTTPS for downloads once curl is rebuilt against OpenSSL.
We rebuild curl with support for OpenSSL.
zlib is a software library used for data compression and implements an abstraction of
DEFLATE algorithm that is also used in gzip.
GNU Automake from 1.16 series that required newer Perl.
GNU Autoconf 2.71 is even newer version of autoconf. It does not build with miniperl, so we postponed it until full perl was built.
Our old patch was built with manual makefile and used mes libc. This is a newer version which we need in order to import gnulib into gettext.
GNU Gettext is an internationalization and localization system used for writing multilingual programs.
Texinfo is a typesetting syntax used for generating documentation. We can now use
makeinfo script to convert .texi files into .info documentation format.
GCC 4.7.4 is the last version written in C. This time we build both C and C++ backends.
The C++ backend has a dependency on gperf, which is written in C++. Fortunately, it is
easy to patch it out; the resulting g++ compiler is capable of building gperf.
We also add in two patchsets to the compiler;
- one to add support for musl shared library support
- one providing a few compiler flags/features that are required later to build GCC 10
This version of binutils provides a more comprehensive set of programming tools for
creating and managing binary programs. It also includes modern versions of the ld
linker, the as assembler and the ar program.
gperf is a perfect hash function generator (hash function is injective).
Library for manipulating Unicode and C strings according to Unicode standard. This is a dependency of GNU Guile.
The libffi library provides a portable, high level programming interface to various calling conventions.
libatomic_ops provides semi-portable access to hardware-provided atomic memory
update operations on a number of architectures.
The Boehm-Demers-Weiser conservative garbage collector can be used as a garbage collecting replacement for C malloc or C++ new.
GNU Ubiquitous Intelligent Language for Extensions (GNU Guile) is the preferred extension language system for the GNU Project and features an implementation of the programming language Scheme.
We use guile-psyntax-bootstrapping project on Guile 3.0.7 to bootstrap
Guile's psyntax.pp without relying on pre-expanded code. This is then
transplanted into Guile 3.0.9.
which shows the full path of (shell) commands. It mostly duplicates
bash built-in command -v but some scripts call which instead.
In particular, autogen scripts use it.
Newer grep will be needed to bootstrap autogen.
Earlier sed was built with manual makefile with most features compiled out.
Build a newer sed using GNU Autotools build system. In particular this will let
sed keep executable bit on after in place editing.
GNU Autogen is a tool designed to simplify the creation and maintenance of
programs that contain large amounts of repetitious text. Unfortunately, the
source is full of pregenerated files that require autogen to rebuild.
We use the gnu-autogen-bootstrapping
project to rebuild those and create (slightly crippled) autogen that
is then able to build a full-featured version.
With GCC and binutils supporting a musl-based toolchain natively, musl itself is rebuilt with support for dynamic linking.
Everything is in place to bootstrap the useful programming language/utility Python. While Python is largely written in C, many parts of the codebase are generated from Python scripts, which only increases as Python matured over time.
We begin with Python 2.0.1, which has minimal generated code, most of which can be removed. Lib/{keyword,token,symbol} scripts are rewritten in C and used to regenerate parts of the standard library. Unicode support and sre (regex) support is stripped out.
Using the stripped-down first version of Python 2.0.1, Python 2.0.1 is rebuilt, including Unicode and regex support (required for future Python builds). The first version is insufficient to run the Lib/{keyword,token,symbol} scripts, so those continue to use the C versions.
Precompiled Python code at this point is highly unreproducible, so it is deleted (JIT compiled instead). This makes Python itself slower, but this is of little consequence.
Python 2.0.1 is sufficient to build Python 2.3.7.
Differences to 2.0.1:
- The new "ast" module, performing parsing of Python, is generated from a parsing specification using Python code.
- 2.0.1 is insufficient to run 2.3.7's unicode regeneration, so Unicode support is again stripped out.
Python 2.3.7 is then rebuilt to include Unicode support.
Python 2.3.7 is sufficient to build Python 2.5.6, with a few minimal changes to language constructs in scripts. This is the last 2.x version we build.
Differences to 2.3.7 are very minimal.
Python 2.5.6 is new enough to be able to build Python 3.1.5, allowing us to move into the modern 3.x series of Python. Various patching is required, as some scripts in the tree are still Python 2 while others are Python 3. We have to convert the Python 3 ones back to Python 2 to be able to use Python 2.5.6.
Differences to 2.5.6:
- An include cycle when a distributed file is removed arises, we have to jump through some hoops to make this work.
- At the second pass of building, various charset encodings can be regenerated & used in the standard library (required in future Python 3.x).
- The new ssl Python library is disabled due to our OpenSSL version being too new.
Python 3.1.5 is rebuilt, using Python 3 for the Python 3 scripts in the tree.
Python 3.1.5 is sufficient to build Python 3.3.7 (rapid language change = small jumps).
Differences to 3.1.5:
- The ssl Python library can now be re-enabled, and
_ssl_data.hregenerated.
Python 3.3.7 is sufficient to build Python 3.4.10.
Differences to 3.3.7:
- The clinic tool has been introduced, which unifies documentation with code. Clinic creates many generated files. We run the clinic tool across all files using clinic.
- The ssl library breaks in much more ugly ways than before, but unlike previous versions, it passes over this error silently.
Python 3.4.10 is sufficient to build Python 3.8.16.
Differences to 3.4.10:
- The build system has been significantly revamped (coming in line with modern standards).
- Many of our previous regenerations can be replaced with one
make regen-allinvocation. - The stringprep Python module, previously deleted, is now required, so it is regenerated.
The newest version of Python, Python 3.11.1 can now be built.
Differences to 3.8.16:
- Unfortunately, the build system has regressed slightly. We must choose the order to perform regenerations in the Makefile ourselves, as some regenerations use other regenerations, but the Makefile does not include them as dependencies.
- The concept of "frozen" modules has been introduced, adding a layer of complexity to regeneration.
stdlib_module_names.his a new file that must be built using data from a current Python binary. To achieve this, a dummystdlib_module_names.his used for the build, thenstdlib_module_names.his created, and Python is rebuilt using the properstdlib_module_names.h. Unfortunately this greatly increases the time taken to build Python, but it is not trivial to work around.- A new generated script
Lib/re/_casefix.pyis introduced. - The ssl module, now unbroken, can be built again.
- Very recent Python versions allow for the use of
SOURCE_DATE_EPOCHto make output of precompiled Python libraries (.pyc) deterministic. Finally, we can re-enable compiling of Python modules.
GCC 10.x series is the last version of GCC that is able to be built with the
C/C++ standards available in GCC 4.7. Instead of manually configuring & compiling
every subdirectory, since we now have autogen available we are able to use
the top-level configure to build the project. We do not use GCC's bootstrap mode,
where GCC is recompiled with itself after being built, since we're just going
to use this GCC to compile GCC 13, it adds build time for little benefit.
We recompile Binutils with the full intended autogen top-level build system, instead of the subdirectory build system used before. This creates a binutils that functions completely correctly for the build of GCC 13 (eg, fixes the mistaken plugin loading support). Other modern features are added, including;
- threaded linking
- 64-bit linking on 32-bit x86
- the modern, rewritten gold linker used by some distributions
This is the most recent version of GCC. With this version of GCC, the final gcc-binutils-musl toolchain is complete. The focus of further builds shifts to rebuilds for correctness, cleanup and preparation for downstream consumption.
In line with this, a variety of modern features + minor build changes are used to ensure the compiler is suitable for downstream consumption;
- A full internal GCC bootstrap is used to ensure there are no lagging historical problems.
- PIE and SSP are enabled by default, as is done on every major modern Linux distribution.
- libssp is disabled and handed off to the libc (done by many modern Linux distributions). libssp in GCC is very broken and glibc-centric - it should really be handled by the libc, which is what most distributions do.
- LTO now fully functions correctly, despite both the linker and the compiler being static binaries.
libmd provides message digest functions. In GNU/Linux, this is typically provided by glibc, but we need libmd to provide it since we are using musl.
libbsd provides BSD-centric functions. We need this in order to build shadow, which expects either glibc or libbsd.
shadow provides a variety of command line utilites to work with users and
groups, avoiding the need for manual modification of /etc/passwd and
/etc/group. This allows unprivileged users to be created by, or for,
post-bootstrap build systems.
opendoas is a port of 'doas' from OpenBSD to Linux. It has all functions of sudo that could be conceivably needed in live-bootstrap, and is much simpler to build. This allows build systems that expect sudo after live-bootstrap to use it.
The version of gzip we have been using up until now is really old, all the way back from mes libc era! Somehow we've managed not to have any problems with it, though. This builds a gzip that is properly packaged and can be handled by all modern build systems.
We already have a perfectly functional diffutils, but some core modern software
does require newer diffutils (understandably, given our diffutils is from 1994).
This also gives the additional diffutils commands diff3 and sdiff.
Similarly to diffutils, our gawk is currently very ancient (1999). That doesn't cut it for modern software such as glibc. We update gawk to a much more modern version.
We are in need of a newer version of m4 for some modern software. Attempts to update m4 1.4.7 earlier in the bootstrap demonstrate some issues with Fiwix, so we build a newer m4 at the end of the bootstrap instead.
Remove any remaining loose build artifacts from steps.
If preseeding was used, this step also removes the repo-preseeded directory.
If FORCE_TIMESTAMPS is enabled, resets all file times in the file system to the Unix epoch, to ensure maximum file system reproducibility.
If checksum updating is enabled, regenerates SHA256SUMS.pkgs to contain the actual hashes of the packages just built.
At the end of the bootstrap, executes any additional shell scripts placed in the
/steps/after directory (if it exists), opens an interactive console (only in
interactive mode), and finally ensures a clean shutdown of the bootstrap system
(only needed in qemu and on bare metal).