Romt (Rust Offline Mirror Tool) aids in using the Rust programming language in an offline context. Instructions and tooling are provided for:
- Mirroring of Rust ecosystem artifacts:
- Toolchains (Rustc, Cargo, libraries, etc.)
- Rustup (toolchain multiplexer)
- Crates.io (community-supplied Crates) with "sparse" index support.
- Incremental artifact downloading (with a configurable number of simultaneous download jobs).
- Incremental artifact transfer to offline network.
- Artifact serving in offline context (offline computer, disconnected network).
Romt support two main mirroring scenarios:
- Development laptop scenario: Download Rust artifacts to the laptop when connected to the Internet, then serve the artifacts from the laptop when offline.
- Disconnected network scenario: Download Rust artifacts on an Internet-connected "Export" machine, transfer them to an offline network, then serve the artifacts from an offline "Import" machine.
Instructions are provided for serving the artifacts using Romt itself via unencrypted HTTP or via the nginx web server.
- Panamax is an alternative to Romt, written in Rust: https://github.com/panamax-rs/panamax
- Python 3.8+ for running
romt
(requires some packages from pypi.org). - Git is required for manipulating the crates.io-index repository.
- Internet-connected computer for initial downloading (Linux, Windows, Mac [1]).
- Offline computer for serving artifacts (Linux, Windows, Mac).
- [Optional] Gnu Privacy Guard (GPG), if installed, is used used for signature checking.
[1] | Note: the author does not have access to a Mac, so support for Romt on Mac is untested but hopefully close to working. Pull requests for Mac-specific fixes are welcome. |
Note
Take note of the instructions for upgrading from Romt versions before 0.4.0 if you have existing crate mirrors created from older Romt versions.
First install prerequisites for Romt:
- Ensure Git is installed; it is required for proper manipulation of the crates.io-index repository.
- For signature checking, GNU Privacy Guard (gpg) should be installed as well.
If gpg is not available, signature files (
*.asc
) will still be transferred but signature checking will be skipped.
Next, choose an option for installation of Romt itself.
The simplest method of installation is to use a pre-built self-contained executable from the Github release area: https://github.com/drmikehenry/romt/releases
Romt is also available in the Python Package Index (PyPI).
Download
romt
source and all dependencies on a host with direct Internet access:Prepare
romt
download area:mkdir romt cd romt
Download poetry and dependencies:
pip download poetry
Download the
romt
source:pip download --no-binary :all: --no-deps romt
Unpack the
romt
source tarballromt-*.tar.gz
:# Example for Linux: tar -zxf romt-*.tar.gz
Download the dependencies from
requirements.txt
:pip download -r romt-*/requirements.txt
If installing to an offline host, transfer the entire
romt/
download area to that host.Ensure that the
PATH
contains the directory that holds installed Python packages:# For Linux: ~/.local/bin # For Windows with Python version X.Y: %APPDATA%\Python\PythonXY\Scripts
Install
romt
from the current directory of sources (ensuring the current working directory is theromt/
download area):pip install --user --no-index --find-links . romt
If desired, the source may be cloned from Github and installed into a virtual environment.
Install Poetry globally as described in the documentation: https://python-poetry.org/docs/#installation
Include the
poetry-plugin-export
plugin as well. Assumingpipx
was used for installation of poetry itself, this is done via:pipx inject poetry poetry-plugin-export
This plugin is needed for generating a
requirements.txt
file.Clone source:
git clone https://github.com/drmikehenry/romt cd romt
Run a Poetry shell (which creates and activates a virtual environment installed with Romt and all dependencies):
poetry shell
Optionally build an executable for your platform:
nox -s build
Find executables in
dist/
tree based on your platform, e.g.:dist/x86_64-linux/romt dist/x86_64-windows/romt.exe dist/aarch64-darwin/romt
Romt is a Python-based command-line tool with several commands:
romt toolchain
: mirror and manage Rust toolchains.romt rustup
: mirror and manage Rustup.romt crate
: mirror and manage crate files from crates.io.romt serve
: simple HTTP server for toolchains, rustup, and crates.
See romt --help
for overall usage help.
In particular, note that romt --readme
will display the contents of this
README file for reference.
For the development-laptop scenario, follow these steps to get a working server configuration with mirrored Rust content.
Ensure the laptop has Internet access.
Install Romt (as above).
Create area for mirrored artifacts:
mkdir mirror cd mirror
Download latest stable toolchain:
# Change ``linux`` to ``windows`` or ``darwin`` as appropriate: romt toolchain -v -s stable -t linux download
Download latest stable rustup version:
# Change ``linux`` to ``windows`` or ``darwin`` as appropriate: romt rustup -v -s stable -t linux download
Setup crate mirror (one-time only):
romt crate init
Download full crates.io mirror:
romt crate -v --keep-going update
Note
A few crates have been removed from crates.io and are therefore not available, so a few download failures (
403 Client Error: Forbidden
) should be expected. The--keep-going
option allows romt to continue in the face of these missing crates.Currently (April 2022), versions of the following crates are missing:
- bork
- css-modules
- css-modules-macros
- deploy
- doccy
- etch
- glib-2-0-sys
- glue
- gobject-2-0-sys
- peek
- pose
Configure crate mirror to be served from localhost (one-time only):
romt crate config
Start Romt as a server on http://localhost:8000:
romt serve
Note
Leave the server running in this dedicated terminal.
Setting up a server for the disconnected-network scenario is similar to that for the development-laptop scenario above; explanations that overlap that scenario are omitted below.
On Internet-connected Export machine:
Install Romt (as above).
Create area for mirrored artifacts:
mkdir mirror cd mirror
Download latest stable toolchain and create
toolchain.tar.gz
:# Change ``linux`` to ``windows`` or ``darwin`` as appropriate: romt toolchain -v -s stable -t linux download pack
Download latest stable rustup version and create
rustup.tar.gz
:# Change ``linux`` to ``windows`` or ``darwin`` as appropriate: romt rustup -v -s stable -t linux download pack
Setup crate mirror (one-time only):
romt crate init
Download and create
crates.tar.gz
:romt crate -v --keep-going export
Transfer
toolchain.tar.gz, ``rustup.tar.gz
, andcrates.tar.gz
to Import machine.
On Disconnected network Import machine:
Install Romt (as above).
Create area for mirrored artifacts (one-time only):
mkdir mirror
Place exported
toolchain.tar.gz, ``rustup.tar.gz
, andcrates.tar.gz
files into thismirror/
directory, and enter the directory at a prompt:cd mirror
Import toolchain and rustup:
romt toolchain -v unpack romt rustup -v unpack
Setup crate mirror (one-time only):
romt crate init-import
Import
crates.tar.gz
:romt crate -v --keep-going import
Configure crate mirror to be served from localhost (one-time only):
romt crate config
Start Romt as a server on http://localhost:8000:
romt serve
Note
Leave the server running in this dedicated terminal.
Follow these steps to configure Rust tooling for use with a mirror server on localhost using either Quick-start server configuration above.
Setup environment variables to point to the server. By default, this will be at http://localhost:8000; adjust all uses of
localhost:8000
below for different server address:port combinations:# For Linux/Mac: export RUSTUP_DIST_SERVER=http://localhost:8000 export RUSTUP_UPDATE_ROOT=http://localhost:8000/rustup # For Windows: set RUSTUP_DIST_SERVER=http://localhost:8000 set RUSTUP_UPDATE_ROOT=http://localhost:8000/rustup
Note
These variables must be set in each terminal window before using the mirror server.
Download the
rustup-init
installer for your platform from the Romt server using the appropriate URL below, saving it into the current directory:Run the installer, accepting the defaults:
# Linux/Mac: chmod +x rustup-init ./rustup-init # Windows rustup-init
Ensure environment changes take place in current shell:
# For Linux/Mac: source $HOME/.cargo/env # For Windows: PATH %USERPROFILE%\.cargo\bin;%PATH%
Try out some rustup commands:
rustup self update rustup component add rust-src
Create the text file
~/.cargo/config.toml
(%USERPROFILE%\.cargo\config.toml
on Windows) to useromt serve
. With a Rust toolchain from 2022-06-20 or later, the "sparse" protocol may be used. This is significantly faster than the older Git-based method.For the sparse index method, use the following contents for the
config.toml
file:[source.crates-io] registry = 'sparse+http://localhost:8000/crates-index/' # Disable cert revocation checking (necessary only on Windows): [http] check-revoke = false
For the older Git-based index method, use the following contents for the
config.toml
file:[source.crates-io] registry = 'http://localhost:8000/git/crates.io-index' # Disable cert revocation checking (necessary only on Windows): [http] check-revoke = false # For greatly improved performance, have Cargo use the Git command-line # client to acquire `crates.io-index` repository. See # https://github.com/rust-lang/cargo/issues/9167 for details. [net] git-fetch-with-cli = true
Create a sample project to demonstrate crate usage:
cargo new rand_test cd rand_test
Add the
rand
crate to the build:cargo add rand
Fetch
rand
and its dependencies:cargo fetch
When upgrading Romt, it's recommended to use the same version of Romt on both the Internet-connected and offline hosts.
Romt 0.4.0 changes how crate files are stored on-disk by default, in order to
fix problems using a mirror with case-sensitive and case-insensitive filesystems
simultaneously. Older Romt stores crates in directories based on the prefix of
each crate's mixed-case name (e.g., MyCrate-0.1.0.crate
would have a prefix
of My/Cr/
). This works for filesystems that are either case-sensitive or
case-insensitive, but it does not allow a tree of crate files created with one
case sensitivity to be accessed using the opposite case sensitivity. Romt 0.4.0
now defaults to making prefix directories in lowercase, allowing a crate mirror
to be used via arbitrary case sensitivity.
For backward compatibility, Romt 0.4.0 supports the use of existing mirror trees
transparently. Newly created mirror trees will use lowercase prefixes by
default (usable on all filesystems); mixed-case prefixes may be requested via
the --prefix=mixed
flag (permitted only with case-sensitive filesystems).
Romt 0.4.0 generates crate archives (crates.tar.gz
) using mixed-case
prefixes by default for backward compatibility, but it can also use lowercase
prefixes for consistency with the preferred on-disk prefix format. To
distinguish the prefix style, Romt 0.4.0 adds an ARCHIVE_FORMAT
file to the
crate archive. Format 1
is compatible with old Romt except for the addition
of the ARCHIVE_FORMAT
file. Old Romt will see this file as an error and
refuse to unpack the archive by default, but processing will succeed using the
invocation romt crate unpack --keep-going
. To avoid corrupting
an existing crate mirror by unpacking a new crate archive with old Romt,
new archives currently default to format 1
, but it's recommended to upgrade
Romt to ensure proper processing of all crate archive formats.
To convert an existing crate mirror (using mixed-case prefixes) to the new format (using lowercase prefixes), the easiest method is to make a crate archive of the old mirror, then unpack the archive using the new format. For example:
# Pack up existing crate mirror into ``crates.tar.gz``:
romt crate -v --keep-going --start 0 --end master pack
# Rename the old crate tree out of the way:
mv crates crates.old
# Initialize for importing with a temporary index area:
romt crate --index index-tmp init-import
# Unpack crates from crates.tar.gz into new crates/ tree:
romt crate -v --index index-tmp unpack
# Verify conversion:
romt crate verify -v --start 0
# Cleanup:
rm -rf index-tmp crates.old
Note that the above steps eliminate the unpredictable-case prefixes that are created with old Romt using a case-insensitive filesystem (such as on Windows).
Romt has some features that are shared across two or more commands.
The TARGET specifies the platform for executables using standard tuple values
(e.g., x86_64-unknown-linux-gnu
). Any tuples supported by Rust are valid.
Typical values are shown below; in parentheses are aliases Romt provides for
ease of typing these common targets:
x86_64-unknown-linux-gnu
(aliaslinux
)x86_64-pc-windows-msvc
(aliaswindows
)aarch64-apple-darwin
(aliasdarwin
)
TARGET values are given by the option --target TARGET
. Multiple TARGET
options may be given, and each TARGET will be split at commas and whitespace to
produce a list of desired TARGET values, e.g.:
--target linux,windows --target 'darwin i686-pc-windows-msvc'
A TARGET may be a literal all
that expands to all known targets. For romt
toolchain
, this list comes from the manifest file. For romt rustup
, it
comes from a hard-code list within Romt; this is an ever-changing list that may
be out-of-date in an old release of Romt.
A TARGET may be a literal *
(asterisk) that expands to all targets with at
least one on-disk file for the given SPEC.
- Each file named
{file}.sha256
contains the SHA256 hash of the corresponding file named{file}
. Romt verifies all hashes to ensure file integrity.
- The option
--num-jobs
controls how many simultaneous download jobs Romt may use at a time. By default,--num-jobs=4
, which should be a conservative value that won't stress the servers heavily. - The option
--timeout
controls the timeout in seconds for downloading. A value of zero disables the timeout functionality altogether. - The option
--assume-ok
instructs Romt that all files already on-disk are to be assumed OK; no hashes or signatures are checked for such files.
The toolchain
operation deals with Rust toolchains.
Each toolchain is identified by a SPEC value which takes on one of the below forms:
{channel} {channel}-{date} {date}
In the above SPEC forms:
{channel}
is typically one of the channel namesnightly
,beta
,stable
. It may also be a version number of the formX.Y.Z
or a literal*
(asterisk) as a wildcard that expands to the setnightly,beta,stable
.{date}
is typically of the formYYYY-MM-DD
(e.g.,2020-04-30
). It may also be a literal*
(asterisk) as a wildcard that expands to all toolchain dates on-disk, or a literallatest
that expands to the most recent toolchain date on-disk.Note that a SPEC value consisting of a single
*
represents a wildcarded{date}
value, not a{channel}
value. It is equivalent to*-*
(making both{channel}
and{date}
wild).Wildcards (
*
andlatest
) may not be used when downloading, and the{channel}
is always required. The{date}
field may be omitted to download the most recent toolchain for the given channel.SPEC values are given by the option
--select SPEC
. Multiple SPEC options may be given, and each SPEC will be split at commas and whitespace to produce a list of desired SPEC values. E.g.:--select nightly,stable --select beta-2020-01-23
See the TARGET section of Commonalities above for details.
A manifest file provides details about a toolchain for a given SPEC, enumerating valid combinations of toolchain components and targets.
The manifest filename is of the form channel-rust-{channel}.toml
, where
{channel}
is one of nightly
, beta
, or stable
. For stable
manifests, the manifest is duplicated into a file of the form
channel-rust-{version}.toml
, where {version}
is a version number of the
form X.Y.Z
.
Downloading is requested via the romt toolchain download
command.
A toolchain is specified by a SPEC/TARGET pair. Both must be given.
Wildcarding (via *
or latest
) is not permitted, though the {date}
may be omitted from the SPEC value, and TARGET may be the literal all
to
download all known targets for the SPEC.
By default, all toolchain components will be downloaded; but when the switch
--cross
is supplied, only the Rust standard library component rust-std
will be downloaded. This is to support cross-compilation to a given target
without the need to download all toolchain components for that target.
Files are downloaded from https://static.rust-lang.org/dist
by default; this
may be changed via the option --url <URL>
.
Files are downloaded to the destination directory dist/
by default; this
may be changed via the option --dest DEST
.
When downloaded, the toolchain will be stored on-disk in the following layout:
dist/ YYYY-MM-DD/ channel-rust-{channel}.toml channel-rust-{channel}.toml.asc channel-rust-{channel}.toml.sha256 {component}-{channel}.tar.xz {component}-{channel}.tar.xz.asc {component}-{channel}.tar.xz.sha256 {component}-{channel}-{target}.tar.xz {component}-{channel}-{target}.tar.xz.asc {component}-{channel}-{target}.tar.xz.sha256
Where:
YYYY-MM-DD
is the toolchain date.{channel}
is one ofnightly
,beta
, orstable
.{component}
represents a toolchain component (e.g.,rust
,cargo
,rust-src
).{target}
represents a target tuple (e.g.,x86_64-unknown-linux-gnu
). Components lacking a{target}
are common across all targets; currently this is limited to therust-src
component.- Each file named
{file}.asc
contains the Gnu Privacy Guard (GPG) digital signature of the corresponding file named{file}
. Checking signature requires GPG; if it is not installed, signature files won't be checked but they will still be transferred. The verification key is available at https://static.rust-lang.org/rust-key.gpg.ascii; this key is built into Romt itself for offline use.
For example, after downloading with this command:
romt toolchain download --select nightly-2020-04-30 --target linux
The tree would contain (among other files):
dist/ 2020-04-30/ channel-rust-nightly.toml channel-rust-nightly.toml.asc channel-rust-nightly.toml.sha256 rust-src-nightly.tar.xz rust-src-nightly.tar.xz.asc rust-src-nightly.tar.xz.sha256 rust-nightly-x86_64-unknown-linux-gnu.tar.xz rust-nightly-x86_64-unknown-linux-gnu.tar.xz.asc rust-nightly-x86_64-unknown-linux-gnu.tar.xz.sha256
For convenience, the most recently released toolchain for each channel
(nightly
, beta
, or stable
) will be copied directly into the
dist/
directory. This is especially helpful for stable
and beta
builds so that the date of the most recent release need not be known in advance.
For stable
manifests, the version-specific copy of the manifest is placed
into dist/
as well.
For example, as of 2020-05-06, the most recent manifests were for SPEC values of:
nightly-2020-05-06
beta-2020-04-26
stable-2020-04-23
(version1.43.0
)
On that date, performing a download with --target linux
and --select
nightly,beta,stable
would yield the following downloaded manifests:
dist/ channel-rust-beta.toml channel-rust-nightly.toml channel-rust-stable.toml channel-rust-1.43.0.toml 2020-04-23/ channel-rust-stable.toml channel-rust-1.43.0.toml 2020-04-26/ channel-rust-beta.toml 2020-05-06/ channel-rust-nightly.toml
Where the dateless manifests housed directly in dist/
are copies of those
from the dated directories.
Because the contents of dateless manifests are subject to change, cached copies
of these files are re-downloaded during a download
command.
Downloaded toolchains may be packed into an ARCHIVE
file using the romt
toolchain pack
command.
The archive file may be moved to another machine and unpacked using the romt
toolchain unpack
command.
Romt will detect when a toolchain has been downloaded via the --cross
switch, in which case only the rust-std
component (along with whatever other
toolchain components are present, if any) will be processed.
For both pack
and unpack
, the ARCHIVE
file is named
toolchain.tar.gz
by default; this may be changed via the option --archive
ARCHIVE
.
An unpack
command automatically performs a verify
(described below). In
addition, dateless manifests are reconstructed automatically during unpack
as part of a fixup operation (described below).
An archive file contains files from dated subdirectories only. Given the
example above for the download
command, the ARCHIVE
would contain only
these manifests:
dist/ 2020-04-23/ channel-rust-stable.toml 2020-04-26/ channel-rust-beta.toml 2020-05-06/ channel-rust-nightly.toml
Each toolchain identified by a SPEC has a canonical manifest file stored in the
toolchain's dated directory. This file has a path of the form
YYYY-MM-DD/channel-rust-{channel}.toml
, where {channel}
is one of the
channel names nightly
, beta
, or stable
.
The "fixup" operation is responsible for making any necessary copies of each
canonical manifest in the dist/
tree. If the given on-disk manifest is
found in the latest dated directory, it will be copied into the top-level
dist/
directory. In addition, for each SPEC on the stable
channel a
version-specific manifest file of the form channel-rust-X.Y.Z.toml
will be
copied into the dated directory and the top-level dist/
directory.
A fixup operation may be explicitly requested via the romt toolchain fixup
command, though that should rarely be required because it is automatically
performed after any download
or unpack
command.
Consider the example above for the download
command; it would generate an
archive containing only these canonical manifests:
dist/ 2020-04-23/ channel-rust-stable.toml 2020-04-26/ channel-rust-beta.toml 2020-05-06/ channel-rust-nightly.toml
The fixup
command would copy these manifests to create:
dist/ channel-rust-beta.toml channel-rust-nightly.toml channel-rust-stable.toml channel-rust-1.43.0.toml 2020-04-23/ channel-rust-stable.toml channel-rust-1.43.0.toml 2020-04-26/ channel-rust-beta.toml 2020-05-06/ channel-rust-nightly.toml
The romt toolchain list
command prints information about on-disk toolchains
for the provided SPEC values. Wildcards are permitted.
For example, the most recent on-disk stable
release can be shown via:
romt toolchain list --select 'stable-latest'
With example output:
stable-2024-02-08(1.76.0) targets[10/94] packages[34/425] x86_64-unknown-linux-gnu native-target x86_64-unknown-linux-musl cross-target
Next to each target name is that target's "type", one of:
native-target
(a full toolchain)cross-target
(a toolchain for cross-compilation)minimal
(minimal toolchain components with no compilation support)
A native-target
contains a full toolchain capable of running on the target
natively (and compiling Rust code to that target as well).
A cross-target
does not contain a compiler that runs on the target; but it
does contain the Rust standard library component rust-std
for the target,
enabling cross-compilation to the target (using a native-target
toolchain on
another host).
A minimal
target lacks rust-std
but has some minimal components
available. Typically a minimal target shows up by coincidence because it shares
one or more components with another target. For example, at the time of this
writing the minimal target mips-unknown-linux-gnu
has no components of its
own in toolchain 1.76.0, but it shares the component rust-docs
with the more
common target x86_64-unknown-linux-gnu
; therefore, downloading the full
toolchain for x86_64-unknown-linux-gnu
will cause mips-unknown-linux-gnu
to be present as a minimal toolchain.
Because most minimal
targets are present only by coincidence and not useful,
listing them is suppressed by default. Use --verbose
to include them,
e.g.:
.. code-block:: sh
romt toolchain list --select 'stable-latest' --verbose
With example output:
List: stable-2024-02-08 [verify] dist/2024-02-08/channel-rust-stable.toml stable-2024-02-08(1.76.0) targets[10/94] packages[34/425] mips-unknown-linux-gnu minimal mips64-unknown-linux-gnuabi64 minimal mips64el-unknown-linux-gnuabi64 minimal mipsel-unknown-linux-gnu minimal mipsisa32r6-unknown-linux-gnu minimal mipsisa32r6el-unknown-linux-gnu minimal mipsisa64r6-unknown-linux-gnuabi64 minimal mipsisa64r6el-unknown-linux-gnuabi64 minimal x86_64-unknown-linux-gnu native-target x86_64-unknown-linux-musl cross-target
To suppress information about targets, use --quiet
:
romt toolchain list --select 'stable-latest' --quiet
With example output:
stable-2024-02-08(1.76.0)
With wildcards, Romt can provide a listing of all available toolchains for a given channel:
romt toolchain list -s 'nightly-*'
With example output:
nightly-2024-02-14(1.78.0) targets[9/94] packages[54/486] x86_64-unknown-linux-gnu native-target nightly-2023-10-31(1.75.0) targets[9/95] packages[54/488] x86_64-unknown-linux-gnu native-target nightly-2023-07-04(1.72.0) targets[5/96] packages[53/529] x86_64-unknown-linux-gnu native-target
After toolchain importation, it may be useful to list toolchains for each channel for reference:
romt toolchain list -s 'nightly-*' > nightly.txt
romt toolchain list -s 'beta-*' > beta.txt
romt toolchain list -s 'stable-*' > stable.txt
For the laptop scenario, only the download
command is needed. After
downloading a toolchain, it will be available for serving via romt serve
(or other means). For example, to download the latest stable toolchain for
Linux:
romt toolchain download --select stable --target linux
For the disconnected network scenario, toolchains are downloaded and packed on an Internet-connected Export machine, then unpacked on an Import machine, e.g.:
On the Export machine:
First, download the latest stable toolchain for Linux into a local
dist/
directory and pack it into an archive for transfer:romt toolchain download pack --select stable --target linux
Transfer the resulting
toolchain.tar.gz
file onto the Import machine.
On the Import machine:
Unpack the archive into a local
dist/
directory:romt toolchain unpack
A few additional commands are provided for romt toolchain
.
romt toolchain fetch-manifest
is the same as download
, but only the
manifest is downloaded.
romt toolchain verify
validates the SHA256 hashes and GPG signatures of
on-disk toolchains. It is implicitly done as part of download
and
unpack
.
romt toolchain all-targets
prints a list of all known targets mentioned in
the given SPEC.
The option --warn-signature
instructs Romt to treat signature failures as
warnings instead of as failures. Signature files will still be downloaded and
transferred. This might be helpful in case the signing key changes.
The option --no-signature
prevents both downloading and checking of GPG
signature files (*.asc
). This is mainly for testing.
The rustup
operation deals with the Rustup toolchain multiplexer.
Each rustup version is identified by a SPEC value which takes on one of the below forms:
{version} stable latest *
In the above SPEC forms:
{version}
is a version number of the formX.Y.Z
.A literal
stable
refers to the current stable version given in therelease-stable.toml
file (described later).A literal
*
(asterisk) is a wildcard that expands to all on-disk versions.A literal
latest
is a wildcard that expands to the latest on-disk version.Wildcards (
*
andlatest
) may not be used when downloading, butstable
is permitted.SPEC values are given by the option
--select SPEC
. Multiple SPEC options may be given, and each SPEC will be split at commas and whitespace to produce a list of desired SPEC values. E.g.:--select stable,1.20.0 --select '1.19.0 1.20.1'
See the TARGET section of Commonalities above for details.
Downloading is requested via the romt rustup download
command.
A rustup executable is specified by a SPEC/TARGET pair. Both must be given.
Wildcarding (via *
or latest
) is not permitted, though SPEC may be the
literal stable
to download the latest stable release, and TARGET may be the
literal all
to download all known targets for the SPEC.
Files are downloaded from https://static.rust-lang.org/rustup
by default;
this may be changed via the option --url <URL>
.
Files are downloaded to the destination directory rustup/
by default;
this may be changed via the option --dest DEST
.
When downloaded, files will be stored on-disk in the following layout:
rustup/ release-stable.toml archive/ {version}/ {target}/ {rustup} {rustup}.sha256 dist/ {target}/
Where:
release-stable.toml
is a configuration file that indicates the most recent stable version of rustup.{version}
is a rustup version of the formX.Y.Z
.{target}
represents a target tuple (e.g.,x86_64-unknown-linux-gnu
).{rustup}
is the name of the rustup executable. On most platforms, this isrustup-init
; on Windows, it'srustup-init.exe
.
For example, if version 1.21.1 were the most recent stable version, after downloading with this command:
romt rustup download --select stable --target linux
The tree would contain:
rustup/ release-stable.toml dist/ x86_64-unknown-linux-gnu/ rustup-init rustup-init.sha256 archive/ 1.21.1/ x86_64-unknown-linux-gnu/ rustup-init rustup-init.sha256
For convenience, all targets found in the most recently released rustup version
will be copied directly into the rustup/dist/
directory.
Because the release-stable.toml
file is subject to change, this file will be
re-downloaded during a download
command when SPEC is stable
.
Downloaded rustup executables may be packed into an ARCHIVE
file using the
romt rustup pack
command.
The archive file may be moved to another machine and unpacked using the
romt rustup unpack
command.
For both pack
and unpack
, the ARCHIVE
file is named
rustup.tar.gz
by default; this may be changed via the option --archive
ARCHIVE
.
An unpack
command automatically performs a verify
(described below). In
addition, the rustup/dist/
tree is created automatically during unpack
as part of a fixup operation (described below).
An archive file contains files from rustup/archive/{version}
subdirectories
only. Given the example above for the download
command, the ARCHIVE
would contain only these files:
rustup/ archive/ 1.21.1/ x86_64-unknown-linux-gnu/ rustup-init rustup-init.sha256
Each rustup version is stored in a directory of the form
rustup/archive/{version}
.
The "fixup" operation is responsible for copying the most recent on-disk
rustup version to rustup/dist/
, and for updating
rustup/release-stable.toml
to contain the most recent version number.
A fixup operation may be explicitly requested via the romt rustup fixup
command, though that should rarely be required because it is automatically
performed after any download
or unpack
command.
Consider the example above for the download
command that generated the
following archive contents:
rustup/ archive/ 1.21.1/ x86_64-unknown-linux-gnu/ rustup-init rustup-init.sha256
Assuming this is the latest on-disk version, the fixup
command would copy
rustup/archive/1.21.1
to rustup/archive
as shown below, and it would
create release-stable.toml
to point to version 1.21.1
:
rustup/ release-stable.toml archive/ 1.21.1/ x86_64-unknown-linux-gnu/ rustup-init rustup-init.sha256 dist/ x86_64-unknown-linux-gnu/ rustup-init rustup-init.sha256
The romt rustup list
command prints information about on-disk rustup
versions for the provided SPEC values. Wildcards are permitted.
For example, the most recent on-disk version can be shown via:
romt rustup list --select 'latest'
With example output:
List: 1.21.1 1.21.1 targets[1] x86_64-unknown-linux-gnu
To suppress information about targets, use --quiet
:
romt rustup list --select 'latest' --quiet
With example output:
1.21.1
With wildcards, Romt can provide a listing of all available rustup versions:
romt rustup list -s '*'
With example output:
List: 1.21.1 1.21.1 targets[1] x86_64-unknown-linux-gnu List: 1.21.0 1.21.0 targets[1] x86_64-unknown-linux-gnu List: 1.20.0 1.20.0 targets[1] x86_64-unknown-linux-gnu
For the laptop scenario, only the download
command is needed. After
downloading a rustup executable, it will be available for serving via romt
serve
(or other means). For example, to download the latest stable rustup for
Linux:
romt rustup download --select stable --target linux
For the disconnected network scenario, rustup versions are downloaded and packed on an Internet-connected Export machine, then unpacked on an Import machine, e.g.:
On the Export machine:
First, download the latest stable rustup for Linux into a local
rustup/
directory and pack it into an archive for transfer:romt rustup download pack --select stable --target linux
Transfer the resulting
rustup.tar.gz
file onto the Import machine.
On the Import machine:
Unpack the archive into a local
rustup/
directory:romt rustup unpack
A few additional commands are provided for romt rustup
.
romt rustup verify
validates the SHA256 hashes of on-disk rustup
executables. It is implicitly done as part of download
and unpack
.
romt rustup all-targets
prints a list of all known targets in Romt's
hard-coded list.
The crate
operation deals with crates (community-written packages of Rust
source code) from the server https://crates.io.
Individual crates are indexed via a Git repository called INDEX. By default,
INDEX is cloned from https://github.com/rust-lang/crates.io-index; this may be
changed with the option --index-url INDEX_URL
.
The INDEX contains one text file for each crate name, where each line of the file is a JSON-formatted description of a single version of that crate. When a new crate file is uploaded, another line is appended to the file and a new commit is made.
The on-disk INDEX directory defaults to git/crates.io-index
; it may be
changed via the option --index INDEX
.
INDEX is essentially a standard Git clone with some additional conventions. It uses the following branches:
remotes/origin/master
The
master
branch of theorigin
repository. Typically this is the repository on Github given by the default value of INDEX_URL.master
The local
master
branch. This is based onremotes/origin/master
, with possible changes to theconfig.json
file (described later).origin_master
A local convenience branch that tracks
remotes/origin/master
. This makes it easy to pushmaster
andremotes/origin/master
to a server.mark
A branch for tracking progress (detailed later).
working
A branch checked out to the working tree and used for merging and modifying repository content; changes are then published atomically to the
master
branch to avoid race conditions.
To keep the number of files in each directory down to a manageable size, the
text files for each crate are distributed into subdirectories based on the first
few characters of the crate's name. The path within INDEX for a crate named
{crate}
is given by {prefix}/{crate}
, where {prefix}
is calculated
based on the length of the crate's name; variations exist for 1-, 2-, 3-, and
4-or-more characters:
{prefix} | crate name length | crate name (as lowercase) |
---|---|---|
1 | 1 | a |
2 | 2 | ab |
3/a | 3 | abc |
ab/cd | 4 or more | abcd* |
The directory names are based on the crate name converted to lowercase so that the repository may be cloned on case-insensitive filesystems (such as on Windows).
For example, the file for the serde
crate would be found by default at
git/crates.io-index/se/rd/serde
.
In addition to per-crate files, there is a config.json
file in the INDEX
that configures the URL for downloading crate files.
A RANGE is defined by a START commit and an END commit. The changes made to the INDEX between START and END represent the list of crates in RANGE that were uploaded to crates.io.
Because START and END represent Git commits, any valid Git commit reference may
be used. In addition, START may be given the value 0
when there is no
starting commit, in which case all commits through END are in RANGE.
The START commit is selected via the option --start START
.
The END commit is selected via the option --end END
.
In general, START and END must both be valid commits in the INDEX; but because
Git branches can't refer to an empty commit, there is no way to initialize a
branch name to a value (like 0
) that means "the start of the repository".
To handle this case, the option --allow-missing-start
indicates that Romt
should treat an unknown branch name for START to be the same as 0
.
Crate files (*.crate
) are tarballs containing Rust source code. Filenames
follow the naming convention {crate}-{version}.crate
, where {crate}
is
the name of the crate (e.g., serde
) and {version}
is the crate's version
number in the form X.Y.Z
.
The URL for a given crate file is given by the template CRATES_URL. The default
value is https://static.crates.io/crates/{crate}/{crate}-{version}.crate; it may
be changed with the option --crates-url CRATES_URL
.
For each crate, the CRATES_URL template will be expanded by replacing
{crate}
with the name of the crate and {version}
with its version. For
example, the default URL for version 1.0.99
of the serde
crate would be:
https://static.crates.io/crates/serde/serde-1.0.99.crate
In addition, {prefix}
and {lowerprefix}
will be replaced with the
crate's prefix and lowercase prefix, respectively (where the construction of the
prefix is explained below).
As an alternative, to use the crate.io API for downloading crates, set CRATES_URL to: https://crates.io/api/v1/crates/{crate}/{version}/download
The INDEX RANGE implies a set of changes (additions, modifications, or removals) made to the INDEX. By default, all crates implied by the RANGE will be used. To restrict to a subset of those crates, crate filters may be used.
A crate filter is of the general form crate_pattern@version_pattern
.
Patterns use file glob syntax (as found in Python's fnmatch
module):
*
matches any sequence of zero or more characters.?
matches any single character.[seq]
matches any character inseq
.[!seq]
matches any character not inseq
.
So, for example:
c*
matchesc
,c2
, andcat
, but not1cat
.c?
matchesc1
andc2
, but notc
orcat
.[ch]*[!g]
matchescat
,hi
, andheat
, but notc
,bat
orbag
.
Sequences may use -
to imply a range; for example, [a-g]
is the same as
[abcdefg]
.
If a pattern is empty, it will be treated as *
.
Operations that apply to the crates in RANGE will be limited by crate filters.
For example, romt crate list --filter mycrate
would list all versions of
the crate named mycrate
, and romt crate list --filter mycrate@0.1.0
would list only version 0.1.0 of mycrate
.
Use --filter FILTER
to supply filter(s) directly on the command line. Use
--filter-file FILTER_FILE
to read filter(s) from one or more FILTER_FILE
files (as if each line were given via --filter
). Both --filter
and
--filter-file
may be given multiple times; their effects aggregate.
A FILTER will be split on runs of spaces, commas, and semi-colons to make it
easier to specify multiple filters in one --filter
switch. For example,
these are equivalent:
romt crate list --filter 'a,b;c,; ,;d' romt crate list --filter a --filter b --filter c --filter d
If a filter contains @
, it will be split into
crate_pattern@version_pattern
components; otherwise, the filter will be used
for the crate_pattern
portion and version_pattern
is implied to be
*
. For example, these pairs are equivalent:
mycrate mycrate@* mycrate@ mycrate@* @ *@* @1.0.? *@1.0.?
Crate filters are applied case-insensitively.
Crate files (*.crate
) are stored on-disk in a directory tree rooted at
CRATES_ROOT, which defaults to crates/
and may be changed via the option
--crates CRATES_ROOT
.
As with the INDEX, crate files are distributed into subdirectories based on the
first few characters of the crate's name. By default, the prefixes are
lowercase (unless forced to mixed-case via romt crate --prefix=mixed
). Romt
versions before 0.4.0 used mixed-case prefixes exclusively, as the author did
not know how to compute lowercase prefixes in nginx rules (this is now solved
using Perl with nginx). Mixed-case prefixes caused problems when accessing a
crates mirror via both case-sensitive and case-insensitive shares
simultaneously, so lowercase prefixes are now preferred.
{prefix} | crate name length | crate name |
---|---|---|
1 | 1 | a |
2 | 2 | ab |
3/a | 3 | abc |
ab/cd | 4 or more | abcd* |
A crate with name {crate}
and version {version}
is found within
CRATES_ROOT at {prefix}/{crate}/{crate}-{version}.crate
.
For example, version 1.0.99 of the serde
crate would be found by default at
crates/se/rd/serde/serde-1.0.99.crate
.
The INDEX and CRATES_ROOT areas must be initialized before use. The initialization method depends on the use.
The romt crate init
command creates the INDEX and CRATES_ROOT areas and
prepares the INDEX as a Git repository with remote named origin
that points
to a Git remote given by INDEX_URL. This is suitable for the laptop scenario
and for the Export machine in the disconnected network scenario.
The romt crate init-import
command is for use on the Import machine in the
disconnected scenario. It's similar to init
, but instead of configuring
INDEX's origin
remote to INDEX_URL, it configures origin
to be a local
bundle file at BUNDLE_PATH that conveys INDEX commits sent from the Export
machine. Subsequent unpack
commands will query the url
key for the
origin
remote within INDEX to determine BUNDLE_PATH. The default value of
BUNDLE_PATH is origin.bundle
within the INDEX directory; this may be changed
via --bundle-path BUNDLE_PATH
.
By default, crate files are stored on-disk using lowercase prefixes. Using
romt crate --prefix=mixed
forces the use of mixed-case prefixes (as used in
Romt before version 0.4.0). Lowercase prefixes are recommended. Romt will not
permit the use of --prefix=mixed
when using case-insensitive filesystems
(such as on Windows) to avoid creating unpredictable-case prefixes due to case
aliasing issues.
Romt (as of version 0.4.0) creates a config.toml
file in CRATES_ROOT as an
implementation detail to aid in the transition to lowercase crate prefixes;
users should generally not have to interact with it. Future versions of Romt
may remove this configuration file and use lowercase prefixes exclusively.
After initialization via init
or init-import
, the local INDEX repository
will be properly setup. If the INDEX contents will be served to clients
directly (e.g., for the laptop scenario or the Import machine in the offline
network scenario), it must be configured for the URL of the offline server by
editing the file config.json
within the top-level directory of INDEX. The
default contents of config.json
(as found on Github) are:
{ "dl": "https://crates.io/api/v1/crates", "api": "https://crates.io" }
The dl
key in particular informs cargo
and other INDEX consumers how to
download crate files cataloged by INDEX.
The romt crate config
command edits config.json
based on the value of
SERVER_URL; this defaults to http://localhost:8000
(as used by romt
serve
, described later). It may be changed via the option --server-url
SERVER_URL
.
Given SERVER_URL, the dl
key will be set to:
SERVER_URL/crates/{crate}/{crate}-{version}.crate
By default, this will be:
http://localhost:8000/crates/{crate}/{crate}-{version}.crate
Rust tooling (e.g., Cargo) will start with the value of the dl
key and
substitute {crate}
with the name of the crate and {version}
with the
crate's version number to form the URL for a given crate file.
Only the SERVER_URL portion of the dl
key is currently configurable; the
rest of the URL is hard-coded to match the conventions of romt serve
.
However, any changes manually committed to config.json
will be preserved by
subsequent Romt operations.
Changes to config.json
are committed to the local working
branch, and
ultimately published to the local master
branch (via the mark
command).
As upstream commits are merged into master
, Romt will ensure that the local
config.json
changes take precedence over possible upstream changes.
Romt uses a branch named mark
as a commit placeholder within INDEX. It
tracks progress through the INDEX, marking one operation's END commit for use as
the next operation's START commit.
The romt crate mark
command sets both the mark
branch and the master
branch to the commit indicated by END. START defaults to mark
such that
subsequent operations pick up where previous ones left off. END defaults to
HEAD
(generally the working
branch) such that RANGE includes all
unprocessed commits.
Note that working copy modifications (merges and edits) are done on the
working
branch. Changes won't be visible on the master
branch until
after the mark
command is executed, ensuring clients won't see partially
complete modifications while the repository is being updated.
Before downloading crate files, the INDEX must be updated. The romt crate
pull
command fetches the latest commits from INDEX's origin
remote into
the remotes/origin/master
branch, then marks this location in the local
branch origin_master
for convenience of reference. The fetched commits are
then merged into the HEAD branch (typically working
), preserving any local
modifications that may have been made to config.json
. If the merge
operation fails, the working copy is reset to remotes/origin/master
and any
local changes to config.json
that may have been present in HEAD
before
the pull are re-applied.
Note: In Romt version 0.1.3 and earlier, HEAD
defaulted to master
,
leaving a small race window where partial modifications to the repository could
be visible to clients (e.g., master
might include mention of a crate that
hasn't yet been downloaded). Therefore, Romt now defaults to using the branch
working
for merging and other modifications to the repository. These
changes won't be visible on master
until the mark
command is invoked.
At each pull
operation, Romt will upgrade the repository to use a
working
branch if HEAD
is not set to working
and the working
branch does not yet exist. To avoid this, pre-create a working
branch (with
arbitrary content) before executing a pull
command, and Romt will not switch
HEAD
to working
.
At times, crates may be removed from the index. If a previously downloaded file is deleted upstream, it may be pruned from the CRATES_ROOT tree (along with any now-empty subdirectories).
The subset of crate files to prune is determined by the RANGE of commits (from START through END) in the INDEX. Each file with a deletion implied by the changes to RANGE will be removed from CRATES_ROOT.
Downloading of crate files is requested via the romt crate download
command.
The subset of crate files to download is determined by the RANGE of commits (from START through END) in the INDEX. Each file is downloaded from the upstream location indicated by CRATES_URL as explained previously. As part of downloading, Romt verifies the SHA256 hash of each crate against the value stored in INDEX to ensure file integrity.
Each crate file is stored below CRATES_ROOT using the prefix mechanism described earlier.
Sometimes individual crate files are removed from the upstream mirror. Romt
warns about such failures and continues with the rest of the crates in the
RANGE. After attempting all crates in RANGE, by default Romt will abort if
any crates failed to download. The option --keep-going
allows Romt to
continue past download failures to subsequent steps (e.g., packing an archive
file).
Use --good-crates GOOD_CRATES
to write the list of "good" crates into the
file GOOD_CRATES. Similarly, use --bad-crates BAD_CRATES
to write the list
of "bad" crates into the file BAD_CRATES. By default, the output for these
files will be of the form crate@version
. With the --show-path
switch,
the .crate
file names will be listed with their relative paths. With the
--show-hash
switch (which implies the --show-path
switch), the
.crate
files will be listed with their SHA256 hashes as well. This is the
same format as provided by romt crate list
; see examples in the section
"Listing crate files".
The romt crate pack
command creates a Git bundle file of the commits in
RANGE, then packs the bundle file along with the downloaded crate files included
in RANGE into an ARCHIVE
file.
The archive file may be moved to another machine and unpacked using the
unpack
command.
For both pack
and unpack
, the ARCHIVE
file is named
crates.tar.gz
by default; this may be changed via the option --archive
ARCHIVE
.
For the pack
command, a Git bundle file is written to disk at BUNDLE_PATH
before being inserted into the ARCHIVE. The default value of BUNDLE_PATH is
origin.bundle
within the INDEX directory; this may be changed via
--bundle-path BUNDLE_PATH
.
An unpack
command extracts the Git bundle file and all crate files, placing
the bundle at the BUNDLE_PATH value specified with the init-import
command.
Crate files are unpacked into CRATES_ROOT. Note that crate files are not
verified automatically as part of the unpack
operation.
An archive file uses the directory structure of CRATES_ROOT for crate files and
the default on-disk location for the Git, and it places the Git bundle file into
the archive with the hard-coded path git/crates.io-index/origin.bundle
. For
example:
git/crates.io-index/origin.bundle crates/3/n/num/num-0.0.1.crate crates/gl/ob/glob/glob-0.0.1.crate crates/se/mv/semver/semver-0.1.0.crate crates/uu/id/uuid/uuid-0.0.1.crate
The romt crate verify
command checks the integrity of each downloaded crate
included in RANGE within INDEX. Using the SHA256 hash values contained in INDEX
for each crate file, Romt ensures that the downloaded crate files have not been
corrupted and that no files in RANGE are missing.
romt crate verify
supports the --good-crates
and --bad-crates
switches in the same way as romt crate download
; see the section
"Downloading" for details.
For each of the three main use cases, there is short command name that implies the needed steps:
update
is the same aspull prune download mark
. This is useful for the laptop scenario.export
is the same aspull prune download pack mark
. This is useful for the Export machine in the disconnected network scenario.import
is the same asunpack pull prune verify mark
. This is useful for the Import machine in the disconnected network scenario.
The romt crate list
command prints the name and version for each crate
included in RANGE within INDEX, independent of whether those crate files have
been downloaded.
For example, to see what new crates are available, first pull
the latest
INDEX and then list
:
romt crate pull list
With example output:
pull... gc@0.3.4 brs@0.2.0 cxx@0.3.1 irc@0.14.0 -scd@0.1.3 [...]
Any crates in the RANGE which have been deleted will be listed with a leading
hyphen; in the example above, scd@0.1.3
has been deleted.
With the --show-path
switch, the .crate
file names will be listed with
their relative paths, e.g.:
romt crate pull list --show-path
With example output:
pull... 2/gc/gc-0.3.4.crate 3/b/brs/brs-0.2.0.crate 3/c/cxx/cxx-0.3.1.crate 3/i/irc/irc-0.14.0.crate -3/s/scd/scd-0.1.3.crate [...]
With the --show-hash
switch (which implies the --show-path
switch), the
.crate
files will be listed with their SHA256 hashes as well, e.g.:
romt crate pull list --show-hash
With example output:
pull... f4917b7233397091baf9136eec3c669c8551b097d69ca2b00a2606e5f07641d1 *2/gc/gc-0.3.4.crate f1e5e58ddd0cfe68b71d5769bec054a98b3adcb3603227b016b2cc6aebee5555 *3/b/brs/brs-0.2.0.crate e2fe8aa3d549e84c89e72a8621281a3f90a6ea771cacf7ed2553f464e49294e0 *3/c/cxx/cxx-0.3.1.crate 245071fa25b5ca1a9995cbc18a5f0bf64e514590525ae96e7d626fe40498440d *3/i/irc/irc-0.14.0.crate -38d847429df942e4db01c64d4119d4d0b9cde270336d2aa4848e80ec8f418b8c *3/s/scd/scd-0.1.3.crate [...]
Because the output is in the format expected by the sha256sum
utility, you
can use the output of romt crate list --show-hash
to verify all crate files
as follows:
romt crate list --start 0 --show-hash > crates.sha256
(cd crates; sha256sum -c ../crates.sha256)
For the laptop scenario, only the update
command is needed, after which
crates will be available for serving via romt serve
(or other means). For
example, to download the latest crates:
romt crate update
For the disconnected network scenario, crate versions are downloaded and packed on an Internet-connected Export machine, then unpacked on an Import machine, e.g.:
On the Export machine:
First, download the latest crates and pack them into
crates.tar.gz
:romt crate export
Transfer the resulting
crates.tar.gz
file onto the Import machine.
On the Import machine:
Unpack the archive:
romt crate import
Prior to Romt version 0.7.0, Romt did not expect crates to be removed or
modified once published; however, crates may in fact be removed from crates.io
in certain circumstances, and after removal a crate may be re-published with a
different hash. With prior versions of Romt, crates deleted from the INDEX were
not deleted from the crates/
directory. Crates deleted from the INDEX
and then re-published with a different hash were not detected, leaving the old
.crate
file in the crates/
directory.
Romt now detects removals and modifications to crates in the INDEX; it will delete
obsolete .crate
files and re-download any modified .crate
file. This
will be done for any crates in the RANGE, but it will not be done retroactively
for crates that were missed in previous updates.
To detect any crates that have been modified and should be re-downloaded, use
the verify
command across all known crates:
romt crate verify --start 0 --bad-crates bad-crates
If bad-crates
contains any crates, re-download and pack just the bad crates
via:
.. code-block:: sh
romt crate download pack --start 0 --filter-file bad-crates
crates.tar.gz
may then be imported as usual.
Any leftover crate files that were removed from INDEX but which remain in
crates/
are harmless and may be left in-place. Should such a crate be
published with a new hash in the future, the new .crate
will automatically
be downloaded and used. If desired, the files may be detected and removed using
Unix command-line tooling (assuming GNU find
and comm
are available) as
follows:
Create a sorted listing of all crate paths currently in the INDEX:
romt crate --start 0 --show-path list | sort > crates-list
Create a sorted listing of all crate paths currently in the
crates/
directory:find crates -name '*.crate' -printf '%P\n' | sort > crates-find
List all paths in
crates/
that are not in the INDEX:comm -23 crates-find crates-list > crates-extra
Examine the contents of
crates-extra
to make sure you want to remove these obsolete files.(optional) Create
crates-extra.tar
containing.crate
files about to be removed:tar -C crates -cf crates-extra.tar -T crates-extra
Remove all files listed in
crates-extra
:cat crates-extra | (cd crates; xargs rm)
The serve
operation runs a local HTTP server exposing toolchain, rustup, and
crate artifacts.
By default, romt serve
listens at the following URL:
http://localhost:8000
To use http://ADDR:PORT
, use the switches --bind ADDR
and/or --port
PORT
.
romt serve
expects the current working directory ($PWD
) to contain all
artifacts being served. Artifacts must be laid out in their default locations
described elsewhere, as follows:
$PWD/ dist/ rustup/ crates/ git/ crates.io-index/
URLs of the form http://ADDR:PORT/{path}
generally map directly to
$PWD/{path}
; exceptions are noted below.
URLs with paths below /crates/
are expected to be of the following form:
http://ADDR:PORT/crates/{crate}/{crate}-{version}.crate
romt serve
will rewrite the URL to insert the expected {prefix}
used in
CRATES_ROOT, effectively transforming the URLs to:
http://ADDR:PORT/crates/{prefix}/{crate}/{crate}-{version}.crate
URLs with paths below /git/
refer to Git repositories. Romt uses
git-http-backend
as distributed with Git to serve these repositories.
For this purpose, romt serve
uses a cgi-bin/
directory in the current
working directory to interface via CGI with git-http-backend
.
Upon launching romt serve
, Romt searches for one of the following files in
cgi-bin/
(depending on the platform):
On Windows:
git-http-backend.bat git-http-backend.exeOn non-Windows:
git-http-backend.sh git-http-backend
If found, Romt will use that file for serving Git repositories via CGI. If not
found, Romt will look in known locations for the git-http-backend
executable
and create a platform-dependent wrapper script in cgi-bin/
to invoke the
executable; the script is named git-http-backend.bat
on Windows and
git-http-backend.sh
on non-Windows.
Currently, Romt probes for the backend in these hard-coded locations (depending on the platform):
- On Windows:
C:/Program Files/Git/mingw64/libexec/git-core/git-http-backend.exe
- On non-Windows:
/usr/lib/git-core/git-http-backend
(typical Linux)/usr/libexec/git-core/git-http-backend
(Alpine Linux)
To manually setup the Git backend, create a script file in cgi-bin/
with
contents similar to these examples (depending on platform):
On Windows, create
cgi-bin/git-http-backend.bat
with contents:@echo off "C:\Program Files\Git\mingw64\libexec\git-core\git-http-backend.exe"
On non-Windows, create
cgi-bin/git-http-backend.sh
with contents:#!/bin/sh exec '/usr/lib/git-core/git-http-backend'
Then make the script executable:
chmod +x cgi-bin/git-http-backend.sh
Rust artifacts may optionally be served via the nginx web server. A simple example for Ubuntu Linux is shown below. If you change host or port values below, configure the index repository via: .. code-block:: sh
romt crate config --server-url <SERVER_URL>
nginx
with Perl support and fcgiwrap
are required. On Ubuntu, these may
be installed via:
apt install nginx-extras fcgiwrap
Below is a sample nginx configuration.
Place the following content into /etc/nginx/sites-available/rust
. Make
adjustments as indicated by each TODO
. These instructions assume crates
are stored using lowercase prefixes; if using mixed-case prefixes, adjust as
directed by the TODO
comments:
server { listen 8000 default_server; listen [::]:8000 default_server; # TODO: Change to absolute path to mirror directory: root /ABSOLUTE/PATH/TO/mirror; server_name _; location / { autoindex on; } # Support serving of Git repositories via git-http-backend. location ~ /git(/.*) { # TODO: Change to absolute path to mirror/git directory: fastcgi_param GIT_PROJECT_ROOT /ABSOLUTE/PATH/TO/mirror/git; include fastcgi_params; fastcgi_pass unix:/var/run/fcgiwrap.socket; fastcgi_param SCRIPT_FILENAME /usr/lib/git-core/git-http-backend; fastcgi_param GIT_HTTP_EXPORT_ALL ""; fastcgi_param PATH_INFO $1; } # Support "sparse" `crates.io-index` protocol. location ~ /crates-index/(.*) { # TODO: Change to absolute path to mirror/git directory: fastcgi_param CRATE_INDEX_ROOT /ABSOLUTE/PATH/TO/mirror/git/crates.io-index; include fastcgi_params; fastcgi_pass unix:/var/run/fcgiwrap.socket; # TODO: Adjust path to `cgi-crates-index` CGI script as needed: fastcgi_param SCRIPT_FILENAME /usr/lib/cgi-bin/cgi-crates-index; fastcgi_param GIT_HTTP_EXPORT_ALL ""; fastcgi_param PATH_INFO $1; } # Rewrite URLs like /crates/{crate}/{crate}-{version}.crate to use # a prefix based on the crate name. Special cases for crate names # with 1, 2, 3, and 4-or-more characters: # a/a-{version}.crate -> 1/a/a-{version}.crate # ab/ab-{version}.crate -> 2/aa/ab-{version}.crate # abc/abc-{version}.crate -> 3/a/abc/abc-{version}.crate # abcd*/abcd*-{version}.crate -> ab/cd/abcd*-{version}.crate # TODO: Comment out this line for mixed-case crate prefixes: rewrite "^/crates/.*$" "$crates_uri" last; # TODO: Uncomment these four lines for mixed-case crate prefixes: # rewrite "^/crates/([^/])/([^/]+)$" "/crates/1/$1/$2" last; # rewrite "^/crates/([^/]{2})/([^/]+)$" "/crates/2/$1/$2" last; # rewrite "^/crates/([^/])([^/]{2})/([^/]+)$" "/crates/3/$1/$1$2/$3" last; # rewrite "^/crates/([^/]{2})([^/]{2})([^/]*)/([^/]+)$" "/crates/$1/$2/$1$2$3/$4" last; }
Serving the crates.io-index
with the "sparse" protocol requires the creation
of the following cgi-crates-index
CGI script. On Ubuntu, such scripts live
in /usr/lib/cgi-bin
; e.g.:
- Create
/usr/lib/cgi-bin
directory if necessary:
mkdir -p /usr/lib/cgi-bin
Create
/usr/lib/cgi-bin/cgi-crates-index
with contents:#!/usr/bin/perl use strict; use warnings; sub send_content { my ($content_type, $body) = @_; my $content_length = length($body); print "Content-Type: $content_type\r\n"; print "Content-Length: $content_length\r\n"; print "\r\n"; print "$body" } sub send_404 { print "Status: 404 Not Found\r\n"; send_content('text/html', <<'END'); <html> <head><title>404 Not Found</title></head> <body> <h1>404 Not Found</h1> </body> </html> END } my $repo = $ENV{CRATE_INDEX_ROOT}; my $path_info = $ENV{PATH_INFO} || "config.json"; my $pipe; if (open($pipe, '-|', "git -C $repo show master:$path_info")) { my $body; { local $/; # Slurp mode. $body = <$pipe>; } if (close($pipe)) { send_content('application/octet-stream', $body); } else { send_404(); } } else { send_404(); }
Note that the user running the CGI script must own the
crates.io-index
tree or else Git may throw errors such as:fatal: detected dubious ownership in repository at '/.../crates.io-index'
Make
/usr/lib/cgi-bin/cgi-crates-index
executable:chmod +x /usr/lib/cgi-bin/cgi-crates-index
Serving crates with lowercase prefixes requires Perl support in nginx (on
Ubuntu, this requires the package nginx-extras
instead of nginx-full
);
Perl support is not required for mixed-case prefixes. To serve crates with
lowercase prefixes, create the file /etc/nginx/conf.d/perl.conf
with the
below contents:
# Reference: https://nginx.org/en/docs/http/ngx_http_perl_module.html # Include the perl module perl_modules perl/lib; # The variable `$crates_uri` will be computed by the Perl subroutine # below, adding a lowercase prefix as required based on the crate name. perl_set $crates_uri 'sub { my $r = shift; my $uri = $r->uri; # Remove all newline characters to avoid CRLF injection vulnerability # (https://stackoverflow.com/questions/3666003/how-i-can-translate-uppercase-to-lowercase-letters-in-a-rewrite-rule-in-nginx-we/68054489#68054489): $uri =~ s/\R//g; if ($uri =~ m@^/crates/([^/])/([^/]+)$@) { $uri = "/crates/1/" . "$1/$2"; } elsif ($uri =~ m@^/crates/([^/]{2})/([^/]+)$@) { $uri = "/crates/2/" . "$1/$2"; } elsif ($uri =~ m@^/crates/([^/])([^/]{2})/([^/]+)$@) { $uri = lc("/crates/3/$1/") . "$1$2/$3"; } elsif ($uri =~ m@^/crates/([^/]{2})([^/]{2})([^/]*)/([^/]+)$@) { $uri = lc("/crates/$1/$2/") . "$1$2$3/$4"; } return $uri; }';
Activate the rust
site via:
ln -s /etc/nginx/sites-available/rust /etc/nginx/sites-enabled/
Currently static artifacts hosted on Rust CDNs are served via Amazon S3 buckets. At times directly accessing the bucket can be helpful.
A helpful command-line tool for use with S3 buckets is awscli
:
https://github.com/aws/aws-cli
Rust https URLs map to S3 bucket URLs as follows:
- https://static.rust-lang.org -> s3://static-rust-lang-org
- https://static.crates.io -> s3://crates-io
Note: unfortunately, the "list" privilege is disabled for the crates-io
bucket.
Here are some common operations on S3 buckets:
List files beginning with PREFIX:
aws s3 ls --no-sign-request s3://BUCKET_NAME/PREFIX
Add
--recursive
flag to recurse into subdirectories.Download a file:
aws s3 cp --no-sign-request s3://BUCKET_NAME/path/file local_file
Examples:
List channel files for toolchain for 2020-04-30:
aws s3 ls --no-sign-request s3://static-rust-lang-org/dist/2020-04-30/chan
with example output:
2020-04-29 20:23:44 10 channel-rust-nightly-date.txt 2020-04-29 20:23:44 833 channel-rust-nightly-date.txt.asc 2020-04-29 20:23:44 96 channel-rust-nightly-date.txt.sha256 2020-04-29 20:23:44 40 channel-rust-nightly-git-commit-hash.txt ...
List
rustup
versions:aws s3 ls --no-sign-request s3://static-rust-lang-org/rustup/archive/
with example output:
PRE 0.2.0/ PRE 0.3.0/ PRE 0.4.0/ ...
Download
serde-1.0.99.crate
:aws s3 cp --no-sign-request s3://crates-io/crates/serde/serde-1.0.99.crate .
This is functionally equivalent to:
curl -O https://static.crates.io/crates/serde/serde-1.0.99.crate
The author has not tested Romt with a proxy server, but user feedback indicates
it's possible (see #10). The
httpx
library's support for proxying is documented at:
https://www.python-httpx.org/advanced/proxies/
httpx
understands several environment variables (documented at
https://www.python-httpx.org/environment_variables/ in the "Proxies" section)
that may be used to influence proxy operation. For example:
HTTP_PROXY
,HTTPS_PROXY
,ALL_PROXY
:Valid values: A URL to a proxy
HTTP_PROXY
,HTTPS_PROXY
,ALL_PROXY
set the proxy to be used for http, https, or all requests respectively.Example:
export HTTP_PROXY=http://my-external-proxy.com:1234 # This request will be sent through the proxy python -c "import httpx; httpx.get('http://example.com')" # This request will be sent directly, as we set `trust_env=False` python -c "import httpx; httpx.get('http://example.com', trust_env=False)"
NO_PROXY
Valid values: a comma-separated list of hostnames/urls
NO_PROXY
disables the proxy for specific urlsExample:
export HTTP_PROXY=http://my-external-proxy.com:1234 export NO_PROXY=http://127.0.0.1,python-httpx.org
In addition, httpx
has information about debugging proxy-related issues at:
https://www.python-httpx.org/contributing/#development-proxy-setup
Also, httpx
can produce more debugging information by setting the
environment variable HTTPX_LOG_LEVEL
to trace
(as documented at
https://www.python-httpx.org/environment_variables/). As a sample invocation on
Linux:
HTTPX_LOG_LEVEL=trace romt toolchain -v -s nightly -t all fetch-manifest
Romt 0.3.0 added support for simultaneous downloading based on the httpx
library; this came with a a default timeout of five seconds which can lead to
ConnectTimeout
or ReadTimeout
errors depending on choice of
--num-jobs
and network characteristics (see
#16).
Romt 0.4.0 adds a --timeout
switch to control this timeout, and changed the
default value to sixty seconds. If timeouts are still occurring, use a larger
timeout value (or use --timeout 0
to disable timeouts altogether).
- "Downloading all the crates on crates.io" provides good reference information on mirroring Rust artifacts: https://www.pietroalbini.org/blog/downloading-crates-io/
- More information on Rust checksumming, signatures, etc., can be found at: https://internals.rust-lang.org/t/future-updates-to-the-rustup-distribution-format/4196
- Information on the "rustup" project: https://github.com/rust-lang/rustup