Bugs and suggestions should be reported at: [yt-dlp/yt-dlp/issues](https://github.com/yt-dlp/yt-dlp/issues). Unless you were prompted to or there is another pertinent reason (e.g. GitHub fails to accept the bug report), please do not send bug reports via personal email. For discussions, join us in our [discord server](https://discord.gg/H5MNcFW63r).
**Please include the full output of yt-dlp when run with `-vU`**, i.e. **add**`-vU` flag to **your command line**, copy the **whole** output and post it in the issue body wrapped in \`\`\` for better formatting. It should look similar to this:
The output (including the first lines) contains important debugging information. Issues without the full output are often not reproducible and therefore will be closed as `incomplete`.
The templates provided for the Issues, should be completed and **not removed**, this helps aide the resolution of the issue.
We often get issue reports that we cannot really decipher. While in most cases we eventually get the required information after asking back multiple times, this poses an unnecessary drain on our resources.
If your report is shorter than two lines, it is almost certainly missing some of these, which makes it hard for us to respond to it. We're often too polite to close the issue outright, but the missing info makes misinterpretation likely. We often get frustrated by these issues, since the only possible way for us to move forward on them is to ask for clarification over and over.
For bug reports, this means that your report should contain the **complete** output of yt-dlp when called with the `-vU` flag. The error message you get for (most) bugs even says so, but you would not believe how many of our bug reports do not contain this information.
If the error is `ERROR: Unable to extract ...` and you cannot reproduce it from multiple countries, add `--write-pages` and upload the `.dump` files you get [somewhere](https://gist.github.com).
**Site support requests must contain an example URL**. An example URL is a URL you might want to download, like `https://www.youtube.com/watch?v=BaW_jenozKc`. There should be an obvious video present. Except under very special circumstances, the main page of a video service (e.g. `https://www.youtube.com/`) is *not* an example URL.
Make sure that someone has not already opened the issue you're trying to open. Search at the top of the window or browse the [GitHub Issues](https://github.com/yt-dlp/yt-dlp/search?type=Issues) of this repository. If there is an issue, subscribe to it to be notified when there is any progress. Unless you have something useful to add to the conversation, please refrain from commenting.
Additionally, it is also helpful to see if the issue has already been documented in the [youtube-dl issue tracker](https://github.com/ytdl-org/youtube-dl/issues). If similar issues have already been reported in youtube-dl (but not in our issue tracker), links to them can be included in your issue report here.
Before requesting a new feature, please have a quick peek at [the list of supported options](README.md#usage-and-options). Many feature requests are for features that actually exist already! Please, absolutely do show off your work in the issue report and detail how the existing similar options do *not* solve your problem.
### Have you read and understood the changes, between youtube-dl and yt-dlp
There are many changes between youtube-dl and yt-dlp [(changes to default behavior)](README.md#differences-in-default-behavior), and some of the options available have a different behaviour in yt-dlp, or have been removed all together [(list of changes to options)](README.md#deprecated-options). Make sure you have read and understand the differences in the options and how this may impact your downloads before opening an issue.
People want to solve problems, and often think they do us a favor by breaking down their larger problems (e.g. wanting to skip already downloaded files) to a specific request (e.g. requesting us to look whether the file exists before downloading the info page). However, what often happens is that they break down the problem into two steps: One simple, and one impossible (or extremely complicated one).
We are then presented with a very complicated request when the original problem could be solved far easier, e.g. by recording the downloaded video IDs in a separate file. To avoid this, you must include the greater context where it is non-obvious. In particular, every feature request that does not consist of adding support for a new site should contain a use case scenario that explains in what situation the missing feature would be useful.
### Does the issue involve one problem, and one problem only?
Some of our users seem to think there is a limit of issues they can or should open. There is no limit of issues they can or should open. While it may seem appealing to be able to dump all your issues into one ticket, that means that someone who solves one of your issues cannot mark the issue as closed. Typically, reporting a bunch of issues leads to the ticket lingering since nobody wants to attack that behemoth, until someone mercifully splits the issue into multiple ones.
In particular, every site support request issue should only pertain to services at one site (generally under a common domain, but always using the same backend technology). Do not request support for vimeo user videos, White house podcasts, and Google Plus pages in the same issue. Also, make sure that you don't post bug reports alongside feature requests. As a rule of thumb, a feature request does not include outputs of yt-dlp that are not immediately related to the feature at hand. Do not post reports of a network error alongside the request for a new video service.
Only post features that you (or an incapacitated friend you can personally talk to) require. Do not post features because they seem like a good idea. If they are really useful, they will be requested by someone who requires them.
Some bug reports are completely unrelated to yt-dlp and relate to a different, or even the reporter's own, application. Please make sure that you are actually using yt-dlp. If you are using a UI for yt-dlp, report the bug to the maintainer of the actual application providing the UI. In general, if you are unable to provide the verbose log, you should not be opening the issue here.
The maintainers and potential contributors of the project often do not have an account for the website you are asking support for. So any developer interested in solving your issue may ask you for account details. It is your personal discretion whether you are willing to share the account in order for the developer to try and solve your issue. However, if you are unwilling or unable to provide details, they obviously cannot work on the issue and it cannot be solved unless some developer who both has an account and is willing/able to contribute decides to solve it.
By sharing an account with anyone, you agree to bear all risks associated with it. The maintainers and yt-dlp can't be held responsible for any misuse of the credentials.
While these steps won't necessarily ensure that no misuse of the account takes place, these are still some good practices to follow.
- Change the password before sharing the account to something random (use [this](https://passwordsgenerator.net/) if you don't have a random password generator).
- Change the password after receiving the account back.
We follow [youtube-dl's policy](https://github.com/ytdl-org/youtube-dl#can-you-add-support-for-this-anime-video-site-or-site-which-shows-current-movies-for-free) to not support services that is primarily used for infringing copyright. Additionally, it has been decided to not to support porn sites that specialize in fakes. We also cannot support any service that serves only [DRM protected content](https://en.wikipedia.org/wiki/Digital_rights_management).
Most users do not need to build yt-dlp and can [download the builds](https://github.com/yt-dlp/yt-dlp/releases), get them via [the other installation methods](README.md#installation) or directly run it using `python -m yt_dlp`.
`yt-dlp` uses [`hatch`](<https://hatch.pypa.io>) as a project management tool.
You can easily install it using [`pipx`](<https://pipx.pypa.io>) via `pipx install hatch`, or else via `pip` or your package manager of choice. Make sure you are using at least version `1.10.0`, otherwise some functionality might not work as expected.
The above command will install a `pre-commit` hook so that required checks/fixes (linting, formatting) will run automatically before each commit. If any code needs to be linted or formatted, then the commit will be blocked and the necessary changes will be made; you should review all edits and re-commit the fixed version.
While it is strongly recommended to use `hatch` for yt-dlp development, if you are unable to do so, alternatively you can manually create a virtual environment and use the following commands:
```shell
# To only install development dependencies:
$ python -m devscripts.install_deps --include dev
# Or, for an editable install plus dev dependencies:
Before you start writing code for implementing a new feature, open an issue explaining your feature request and at least one use case. This allows the maintainers to decide whether such a feature is desired for the project in the first place, and will provide an avenue to discuss some implementation details. If you open a pull request for a new feature without discussing with us first, do not be surprised when we ask for large changes to the code, or even reject it outright.
If you want to add support for a new site, first of all **make sure** this site is **not dedicated to [copyright infringement](#is-the-website-primarily-used-for-piracy)**. yt-dlp does **not support** such sites thus pull requests adding support for them **will be rejected**.
After you have ensured this site is distributing its content legally, you can follow this quick list (assuming your service is called `yourextractor`):
1. Add an import in [`yt_dlp/extractor/_extractors.py`](yt_dlp/extractor/_extractors.py). Note that the class name must end with `IE`. Also note that when adding a parenthesized import group, the last import in the group must have a trailing comma in order for this formatting to be respected by our code formatter.
1. Run `hatch test YourExtractor`. This *may fail* at first, but you can continually re-run it until you're done. Upon failure, it will output the missing fields and/or correct values which you can copy. If you decide to add more than one test, the tests will then be named `YourExtractor`, `YourExtractor_1`, `YourExtractor_2`, etc. Note that tests with an `only_matching` key in the test's dict are not included in the count. You can also run all the tests in one go with `YourExtractor_all`
1. Make sure you have at least one test for your extractor. Even if all videos covered by the extractor are expected to be inaccessible for automated testing, tests should still be added with a `skip` parameter indicating why the particular test is disabled from running.
1. Have a look at [`yt_dlp/extractor/common.py`](yt_dlp/extractor/common.py) for possible helper methods and a [detailed description of what your extractor should and may return](yt_dlp/extractor/common.py#L119-L440). Add tests and code for as many as you want.
1. Make sure your code follows [yt-dlp coding conventions](#yt-dlp-coding-conventions), passes [ruff](https://docs.astral.sh/ruff/tutorial/#getting-started) code checks and is properly formatted:
You can use `hatch fmt` to automatically fix problems. Rules that the linter/formatter enforces should not be disabled with `# noqa` unless a maintainer requests it. The only exception allowed is for old/printf-style string formatting in GraphQL query templates (use `# noqa: UP031`).
1. Make sure your code works under all [Python](https://www.python.org/) versions supported by yt-dlp, namely CPython and PyPy for Python 3.8 and above. Backward compatibility is not required for even older versions of Python.
1. When the tests pass, [add](https://git-scm.com/docs/git-add) the new files, [commit](https://git-scm.com/docs/git-commit) them and [push](https://git-scm.com/docs/git-push) the result, like this:
**Tip:** To test extractors that require login information, create a file `test/local_parameters.json` and add `"usenetrc": true` or your `username`&`password` or `cookiefile`/`cookiesfrombrowser` in it:
Extractors are very fragile by nature since they depend on the layout of the source data provided by 3rd party media hosters out of your control and this layout tends to change. As an extractor implementer your task is not only to write code that will extract media links and metadata correctly but also to minimize dependency on the source's layout and even to make the code foresee potential future changes and be ready for that. This is important because it will allow the extractor not to break on minor layout changes thus keeping old yt-dlp versions working. Even though this breakage issue may be easily fixed by a new version of yt-dlp, this could take some time, during which the extractor will remain broken.
For extraction to work yt-dlp relies on metadata your extractor extracts and provides to yt-dlp expressed by an [information dictionary](yt_dlp/extractor/common.py#L119-L440) or simply *info dict*. Only the following meta fields in the *info dict* are considered mandatory for a successful extraction process by yt-dlp:
The aforementioned metafields are the critical data that the extraction does not make any sense without and if any of them fail to be extracted then the extractor is considered completely broken. While all extractors must return a `title`, they must also allow it's extraction to be non-fatal.
The extractor is allowed to return the info dict without url or formats in some special cases if it allows the user to extract useful information with `--ignore-no-formats-error` - e.g. when the video is a live stream that has not started yet.
[Any field](yt_dlp/extractor/common.py#219-L426) apart from the aforementioned ones are considered **optional**. That means that extraction should be **tolerant** to situations when sources for these fields can potentially be unavailable (even if they are always available at the moment) and **future-proof** in order not to break the extraction of general purpose mandatory fields.
Assume you want to extract `summary` and put it into the resulting info dict as `description`. Since `description` is an optional meta field you should be ready that this key may be missing from the `meta` dict, so that you should extract it like:
The latter will break extraction process with `KeyError` if `summary` disappears from `meta` at some later time but with the former approach extraction will just go ahead with `description` set to `None` which is perfectly fine (remember `None` is equivalent to the absence of data).
Similarly, you should pass `fatal=False` when extracting optional data from a webpage with `_search_regex`, `_html_search_regex` or similar methods, for instance:
```python
description = self._search_regex(
r'<span[^>]+id="title"[^>]*>([^<]+)<',
webpage, 'description', fatal=False)
```
With `fatal` set to `False` if `_search_regex` fails to extract `description` it will emit a warning and continue extraction.
You can also pass `default=<some fallback value>`, for example:
On failure this code will silently continue the extraction with `description` set to `None`. That is useful for metafields that may or may not be present.
In this case, `thumbnail_data` will be `None` if the field was not found and this will cause the loop `for item in thumbnail_data` to raise a fatal error. Using `or []` avoids this error and results in setting an empty list in `thumbnails` instead.
When extracting metadata try to do so from multiple sources. For example if `title` is present in several places, try extracting from at least some of them. This makes it more future-proof in case some of the sources become unavailable.
Assume that you have some another source you can extract `title` from, for example `og:title` HTML meta of a `webpage`. In this case you can provide a fallback like:
Capturing group must be an indication that it's used somewhere in the code. Any group that is not used must be non capturing.
##### Example
Don't capture id attribute name here since you can't use it for anything anyway.
Correct:
```python
r'(?:id|ID)=(?P<id>\d+)'
```
Incorrect:
```python
r'(id|ID)=(?P<id>\d+)'
```
#### Make regular expressions relaxed and flexible
When using regular expressions try to write them fuzzy, relaxed and flexible, skipping insignificant parts that are more likely to change, allowing both single and double quotes for quoted values and so on.
#### Keep the regular expressions as simple as possible, but no simpler
Since many extractors deal with unstructured data provided by websites, we will often need to use very complex regular expressions. You should try to use the *simplest* regex that can accomplish what you want. In other words, each part of the regex must have a reason for existing. If you can take out a symbol and the functionality does not change, the symbol should not be there.
#### Do not misuse `.` and use the correct quantifiers (`+*?`)
Avoid creating regexes that over-match because of wrong use of quantifiers. Also try to avoid non-greedy matching (`?`) where possible since they could easily result in [catastrophic backtracking](https://www.regular-expressions.info/catastrophic.html)
Correct:
```python
title = self._search_regex(r'<span\b[^>]+class="title"[^>]*>([^<]+)', webpage, 'title')
```
Incorrect:
```python
title = self._search_regex(r'<span\b.*class="title".*>(.+?)<', webpage, 'title')
There is a soft limit to keep lines of code under 100 characters long. This means it should be respected if possible and if it does not make readability and code maintenance worse. Sometimes, it may be reasonable to go upto 120 characters and sometimes even 80 can be unreadable. Keep in mind that this is not a hard limit and is just one of many tools to make the code more readable.
Conversely, don't unnecessarily split small lines further. As a rule of thumb, if removing the line split keeps the code under 80 characters, it should be a single line.
Always use single quotes for strings (even if the string has `'`) and double quotes for docstrings. Use `'''` only for multi-line strings. An exception can be made if a string has multiple single quotes in it and escaping makes it *significantly* harder to read. For f-strings, use you can use double quotes on the inside. But avoid f-strings that have too many quotes inside.
Extracting variables is acceptable for reducing code duplication and improving readability of complex expressions. However, you should avoid extracting variables used only once and moving them to opposite parts of the extractor file, which makes reading the linear flow difficult.
Always move trailing parentheses used for grouping/functions after the last argument. On the other hand, multi-line literal list/tuple/dict/set should closed be in a new line. Generators and list/dict comprehensions may use either style
Wrap all extracted numeric data into safe functions from [`yt_dlp/utils/`](yt_dlp/utils/): `int_or_none`, `float_or_none`. Use them for string to number conversions as well.
Use `unified_strdate` for uniform `upload_date` or any `YYYYMMDD` meta field extraction, `unified_timestamp` for uniform `timestamp` extraction, `parse_filesize` for `filesize` extraction, `parse_count` for count meta fields extraction, `parse_resolution`, `parse_duration` for `duration` extraction, `parse_age_limit` for `age_limit` extraction.
The `pending-fixes` label is added when there are changes requested to a PR. When the necessary changes are made, the label should be removed. However, despite our best efforts, it may sometimes happen that the maintainer did not see the changes or forgot to remove the label. If your PR is still marked as `pending-fixes` a few days after all requested changes have been made, feel free to ping the maintainer who labeled your issue and ask them to re-review and remove the label.
