The curldl Python module safely and reliably downloads files with PycURL, which in turn is a wrapper for libcurl file transfer library. The purpose of curldl is providing a straightforward API for downloading files with the following features:
- Multi-protocol support: protocol support is delegated to curl in as protocol-neutral way as possible. This means that there is no reliance on HTTP-specific header and statuses, for example. If a feature like download resuming and if-modified-since condition is supported by the underlying protocol, it can be used by curldl.
- If a partial download is abandoned, most chances are that it may be resumed later (supported for HTTP(S), FTP(S) and FILE protocols). A
.partextension is added to the partial download file, and it is renamed to the target file name once the download completes. - If the downloaded file exists, it is not downloaded again unless the timestamp of file on server is newer (supported for HTTP(S), FTP(S), RTSP and FILE protocols). Note that file download may be skipped before timestamp is considered due to file size matching the expected file size (see below).
- Downloads are configured relative to a base directory, and relative download path is verified not to escape the base directory directly, via symlinks, or otherwise.
- Downloaded file size and/or cryptographic digest(s) can be verified upon download completion. This verification, together with the relative path safety above, allows for easy implementation of mirroring scripts — e.g., when relative path, file size and digest are located in a downloaded XML file.
- Speed: since native libcurl writes directly to the output stream file descriptor, there are no transfers of large chunks of data inside Python interpreter.
Most examples below use the curldl wrapper script instead of Python code. Of course, in all cases it is easy to write a few lines of code with identical functionality — see the first example. Also, note that inline documentation is available for all functions.
The following code snippet downloads a file and verifies its size and SHA-1 digest. A progress bar is shown on stderr while download is in progress.
import curldl, os
dl = curldl.Curldl(basedir="downloads", progress=True)
dl.get("https://kernel.org/pub/linux/kernel/Historic/linux-0.01.tar.gz",
"linux-0.01.tar.gz", size=73091,
digests={"sha1": "566b6fb6365e25f47b972efa1506932b87d3ca7d"})
assert os.path.exists("downloads/linux-0.01.tar.gz")If verification fails, the partial download is removed; otherwise it is renamed to the target file after being timestamped with last-modified timestamp received from the server.
A similar result is achieved on command-line by running the CLI wrapper script, which is useful for quickly testing curldl functionality:
curldl -b downloads -s 73091 -a sha1 -d 566b6fb6365e25f47b972efa1506932b87d3ca7d \
-p -l debug https://kernel.org/pub/linux/kernel/Historic/linux-0.01.tar.gzThe corresponding (redacted) log output:
Saving download(s) to: linux-0.01.tar.gz
Creating directory: downloads
Downloading https://kernel.org/pub/linux/kernel/Historic/linux-0.01.tar.gz
to downloads/linux-0.01.tar.gz.part
Finished downloading downloads/linux-0.01.tar.gz.part 0 -> 73,091 B
(HTTPS 200: OK) [0:00:01]
Timestamping downloads/linux-0.01.tar.gz.part with 1993-10-30 00:00:00+00:00
Successfully verified file size of downloads/linux-0.01.tar.gz.part
Successfully verified SHA1 of downloads/linux-0.01.tar.gz.part
Partial download of downloads/linux-0.01.tar.gz passed verification
(73091 / {'sha1': '566b6fb6365e25f47b972efa1506932b87d3ca7d'})
Moving downloads/linux-0.01.tar.gz.part to downloads/linux-0.01.tar.gz
Note that renaming of downloads/linux-0.01.tar.gz.part to downloads/linux-0.01.tar.gz is the very last action of Curldl.get() method. If the target filename exists, the download succeeded and passed verification, if requested.
Running the same command again doesn't actually result in a server request since file size matches (digest is not checked since it would be time-prohibitive when mirroring large repositories):
Saving download(s) to: linux-0.01.tar.gz
Skipping update of downloads/linux-0.01.tar.gz since it has
the expected size 73,091 B
We can also request the same file without providing an expected size:
import curldl
dl = curldl.Curldl(basedir="downloads", progress=True)
dl.get("ftp://ftp.hosteurope.de/mirror/ftp.kernel.org/pub/linux/kernel/Historic/linux-0.01.tar.gz",
"linux-0.01.tar.gz")In this case, the download is skipped due to If-Modified-Since check:
Will update downloads/linux-0.01.tar.gz.part
if modified since 1993-10-30 00:00:00+00:00
Discarding downloads/linux-0.01.tar.gz.part because
it is not more recent
Note that FTP protocol was used this time — curldl is entirely protocol-agnostic when using the underlying libcurl functionality.
If a download is interrupted, it will be resumed on the next attempt (which may also be a retry according to the configured retry policy). Here is what happens when Ctrl-C is used to send a SIGINT signal to the Python process. This example also demonstrates how to construct a filename from a URL (CLI interface does the same when --output switch is omitted).
import curldl, os, urllib.parse as uparse
dl = curldl.Curldl(basedir="downloads", progress=True)
url = "https://releases.ubuntu.com/22.04.2/ubuntu-22.04.2-live-server-amd64.iso"
filename = os.path.basename(uparse.unquote(uparse.urlparse(url).path))
dl.get(url, filename)The corresponding (redacted) log output:
Downloading https://releases.ubuntu.com/22.04.2/ubuntu-22.04.2-live-server-amd64.iso
to downloads/ubuntu-22.04.2-live-server-amd64.iso.part
ubuntu-22.04.2-live-server-amd64.iso: 13%|██▋ | 244M/1.84G
[00:06<00:38, 44.7MB/s] ^C KeyboardInterrupt:
Download interrupted downloads/ubuntu-22.04.2-live-server-amd64.iso.part 0 -> 259,981,312 B
(42: Callback aborted / HTTPS 200: OK) [0:00:07]
Attempting the same download again resumes the download:
Resuming download of https://releases.ubuntu.com/22.04.2/ubuntu-22.04.2-live-server-amd64.iso
to downloads/ubuntu-22.04.2-live-server-amd64.iso.part at 259,981,312 B
Finished downloading downloads/ubuntu-22.04.2-live-server-amd64.iso.part
259,981,312 -> 1,975,971,840 B (HTTPS 206: Partial Content) [0:01:24]
Note, however, that we didn't provide a size or digest for verification. Since the downloaded file is timestamped only once download completes, how does curldl know that the file wasn't changed on the server in the meantime? The answer is that curldl simply avoids removing large partial downloads in such cases — see documentation for always_keep_part_bytes constructor parameter of Curldl.
By default, curldl enables the following protocols:
- HTTP(S)
- FTP(S)
- SFTP
In order to enable a different set of protocols, use the allowed_protocols_bitmask constructor argument. For instance, the code below downloads a file:// URI:
import curldl, pycurl, pathlib
protocols = pycurl.PROTO_FILE | pycurl.PROTO_HTTPS
dl = curldl.Curldl(basedir="downloads", allowed_protocols_bitmask=protocols)
file_uri = pathlib.Path(__file__).absolute().as_uri()
dl.get(file_uri, "current_source.py")To enable all protocols, use allowed_protocols_bitmask=pycurl.PROTO_ALL. Note, however, that there might be security repercussions.
Attempts to escape base directory are prevented, e.g.:
import curldl, os
dl = curldl.Curldl(basedir=os.curdir)
dl.get("http://example.com/", os.path.join(os.pardir, "file.txt"))The above results in:
ValueError: Relative path ../file.txt escapes base path /home/user/curldl
curldl performs extensive checks to prevent escaping the base download directory — see FileSystem class implementation and unit tests for details.
The only requirement for curldl is Python 3.8+. Install the package as follows:
pip install curldlIf you encounter a build failure during installation of pycurl dependency, the following should help:
- On Linux, install one of:
- pycurl package from distribution repo — e.g., on Ubuntu run
sudo apt install python3-pycurl - libcurl development files with
sudo apt install build-essential libcurl4-openssl-dev
- pycurl package from distribution repo — e.g., on Ubuntu run
- On Windows, install an unofficial pycurl build since official builds are not available at the moment — e.g., from Christoph Gohlke's packages, or use Conda (see below).
- On Windows and macOS, use Conda or Miniconda with conda-forge channel. For instance, see runtime dependencies in the following test environment.
Overall, curldl is expected to not have any issues in any environment with Python 3.8+ (CPython or PyPy) — see the Testing section below.
A simplified configuration matrix covered by CI/CD test + build pipeline is presented below:
| Platform | 3.8 | 3.9 | 3.10 | 3.11 | 3.12 | 3.13 | PyPy 3.8 | PyPy 3.10 |
|---|---|---|---|---|---|---|---|---|
| Ubuntu-x64 | v c | v | v p | v c | v c p | v | v | v |
| Windows-x64 | v c | v | v c | v c | ||||
| Windows-x86 | v | v | v | v | ||||
| macOS-x64 | c | c | c |
In the table:
- v (venv) — virtual environment with all package dependencies and editable package install; on Ubuntu includes tests with minimal versions of package dependencies;
- c (conda) — Miniconda with package dependencies installed from mini-forge channel, and curldl as editable package install;
- p (platform) — as many dependencies as possible satisfied via Ubuntu package repository, and curldl as wheel install; see images for OS and platform Python versions.
The CI/CD pipeline succeeds only if curldl package successfully builds and passes all the pytest test cases with 100% code coverage, as well as Pylint, Mypy and Bandit static code analysis. Code style checks are also a part of the pipeline. Note that the testing code is also covered by these restrictions.
The following commands install and activate a venv environment in Linux, using the available Python 3 interpreter:
./venv.sh install-venv
. venv/bin/activatevenv.sh is a convenience venv wrapper that also enables some additional Python checks; you can use it to run Python code, or just activate the venv environment instead as shown above.
Use pytest in order to run all test cases:
./venv.sh pytestIn addition to the actual tests, pytest executes Pylint, Mypy, code coverage and code formatting plugins (black and isort). Thus, make sure that all new code is covered by tests.
Testing with Conda is possible as well — see the CI/CD pipeline execution for details.
Reformat the code with black and isort by running the following scripts:
misc/scripts/run-black.sh
misc/scripts/run-isort.shThis is only necessary if the tests fail due to code formatting.
Upon authoring a set of code or documentation changes, prepare a changelog fragment using towncrier as follows:
towncrier create -c "Extend package usage documentation" 54.doc.mdThe command above creates a changelog fragment file in docs/changelog.d. See the output of towncrier create --help for supported fragment types. Note also that 54 above is the number of GitHub pull request, which is used to format a pull request link when generating the combined changelog. So creation of new fragment needs to be done after opening a pull request.
When releasing a new version, the changelog file can be updated as follows:
towncrier build --version 1.x.y [--draft]Add --draft option for a dry run first, because otherwise fragment files will be removed, and changelog file extended with the new entries.
See the Changelog file for a summary of changes in each release.
This project is released under the GNU LGPL License Version 3 or any later version.
