| (md/security)= |
| |
| # Security |
| |
| Many people don't understand that markdown format does not care much about security. |
| In many cases you have to pass output to sanitizers. |
| `markdown-it` provides 2 possible strategies to produce safe output: |
| |
| 1. Don't enable HTML. Extend markup features with [plugins](md/plugins). |
| We think it's the best choice and use it by default. |
| - That's ok for 99% of user needs. |
| - Output will be safe without sanitizer. |
| 2. Enable HTML and use external sanitizer package(s). |
| |
| Also by default `markdown-it` prohibits some kind of links, which could be used |
| for XSS: |
| |
| - `javascript:`, `vbscript:` |
| - `file:` |
| - `data:`, except some images (gif/png/jpeg/webp). |
| |
| So, by default `markdown-it` should be safe. We care about it. |
| |
| If you find a security problem - contact us via <executablebooks@gmail.com>. |
| Such reports are fixed with top priority. |
| |
| ## Plugins |
| |
| Usually, plugins operate with tokenized content, and that's enough to provide safe output. |
| |
| But there is one non-evident case you should know - don't allow plugins to generate arbitrary element `id` and `name`. |
| If those depend on user input - always add prefixes to avoid DOM clobbering. |
| See [discussion](https://github.com/markdown-it/markdown-it/issues/28) for details. |
| |
| So, if you decide to use plugins that add extended class syntax or autogenerating header anchors - be careful. |
| |
| (md/performance)= |
| |
| # Performance |
| |
| You can view our continuous integration benchmarking analysis at: <https://executablebooks.github.io/markdown-it-py/dev/bench/>, |
| or you can run it for yourself within the repository: |
| |
| ```console |
| $ tox -e py38-bench-packages -- --benchmark-columns mean,stddev |
| |
| Name (time in ms) Mean StdDev |
| --------------------------------------------------------------- |
| test_mistune 70.3272 (1.0) 0.7978 (1.0) |
| test_mistletoe 116.0919 (1.65) 6.2870 (7.88) |
| test_markdown_it_py 152.9022 (2.17) 4.2988 (5.39) |
| test_commonmark_py 326.9506 (4.65) 15.8084 (19.81) |
| test_pymarkdown 368.2712 (5.24) 7.5906 (9.51) |
| test_pymarkdown_extra 640.4913 (9.11) 15.1769 (19.02) |
| test_panflute 678.3547 (9.65) 9.4622 (11.86) |
| --------------------------------------------------------------- |
| ``` |
| |
| As you can see, `markdown-it-py` doesn't pay with speed for it's flexibility. |
| |
| ```{note} |
| `mistune` is not CommonMark compliant, which is what allows for its |
| faster parsing, at the expense of issues, for example, with nested inline parsing. |
| See [mistletoes's explanation](https://github.com/miyuchina/mistletoe/blob/master/performance.md) |
| for further details. |
| ``` |