@title Rendering HTML @group developer Rendering HTML in the Phabricator environment. = Overview = Phabricator attempts to prevent XSS by treating strings as default-unsafe when rendering. This means that if you try to build HTML through string concatenation, it won't work: the string will be escaped by the rendering pipeline, and the browser will treat it as plain text, not HTML. This document describes the right way to build HTML components so they are safe from XSS and render correctly. Broadly: - Use @{function@libphutil:phutil_tag} (and @{function:javelin_tag}) to build tags. - Use @{function@libphutil:hsprintf} where @{function@libphutil:phutil_tag} is awkward. - Combine elements with arrays, not string concatenation. - @{class:AphrontView} subclasses should return a @{class@libphutil:PhutilSafeHTML} object from their `render()` method. - @{class:AphrontView} subclasses act like tags when rendering. - @{function:pht} has some special rules. - There are some other things that you should be aware of. See below for discussion. = Building Tags: phutil_tag() = Build HTML tags with @{function@libphutil:phutil_tag}. For example: phutil_tag( 'div', array( 'class' => 'some-class', ), $content); @{function@libphutil:phutil_tag} will properly escape the content and all the attributes, and return a @{class@libphutil:PhutilSafeHTML} object. The rendering pipeline knows that this object represents a properly escaped HTML tag. This allows @{function@libphutil:phutil_tag} to render tags with other tags as content correctly (without double-escaping): phutil_tag( 'div', array(), phutil_tag( 'strong', array(), $content)); In Phabricator, the @{function:javelin_tag} function is similar to @{function@libphutil:phutil_tag}, but provides special handling for the `sigil` and `meta` attributes. = Building Blocks: hsprintf() = Sometimes, @{function@libphutil:phutil_tag} can be particularly awkward to use. You can use @{function@libphutil:hsprintf} to build larger and more complex blocks of HTML, when @{function@libphutil:phutil_tag} is a poor fit. @{function:hsprintf} has `sprintf()` semantics, but `%s` escapes HTML: // Safely build fragments or unwieldy blocks. hsprintf( '
', $div_id); @{function:hsprintf} can be especially useful when: - You need to build a block with a lot of tags, like a table with rows and cells. - You need to build part of a tag (usually you should avoid this, but if you do need to, @{function@libphutil:phutil_tag} can not do it). Note that it is unsafe to provide any user-controlled data to the first parameter of @{function@libphutil:hsprintf} (the `sprintf()`-style pattern). Like @{function@libphutil:phutil_tag}, this function returns a @{class@libphutil:PhutilSafeHTML} object. = Composing Tags = When you are building a view which combines smaller components, like a section with a header and a body: $header = phutil_tag('h1', ...); $body = phutil_tag('p', ...); ...you should NOT use string concatenation: COUNTEREXAMPLE // Not dangerous, but does the wrong thing. phutil_tag('div', array(), $header.$body); Instead, use an array: // Render a tag containing other tags safely. phutil_tag('div', array(), array($header, $body)); If you concatenate @{class@libphutil:PhutilSafeHTML} objects, they revert to normal strings and are no longer marked as properly escaped tags. (In the future, these objects may stop converting to strings, but for now they must to maintain backward compatibility.) If you need to build a list of items with some element in between each of them (like a middot, comma, or vertical bar) you can use @{function:phutil_implode_html}: // Render links with commas between them. phutil_tag( 'div', array(), phutil_implode_html(', ', $list_of_links)); = AphrontView Classes = Subclasses of @{class:AphrontView} in Phabricator should return a @{class@libphutil:PhutilSafeHTML} object. The easiest way to do this is to return `phutil_tag()` or `javelin_tag()`: return phutil_tag('div', ...); You can use an @{class:AphrontView} subclass like you would a tag: phutil_tag('div', array(), $view); = Internationalization: pht() = The @{function:pht} function has some special rules. If any input to @{function:pht} is a @{class@libphutil:PhutilSafeHTML} object, @{function:pht} returns a @{class@libphutil:PhutilSafeHTML} object itself. Otherwise, it returns normal text. This is generally safe because translations are not permitted to have more tags than the original text did (so if the original text had no tags, translations can not add any). Normally, this just means that @{function:pht} does the right thing and behaves like you would expect, but it is worth being aware of. = Special Cases = NOTE: This section describes dangerous methods which can bypass XSS protections. If possible, do not use them. You can build @{class@libphutil:PhutilSafeHTML} out of a string explicitly by calling @{function:phutil_safe_html} on it. This is **dangerous**, because if you are wrong and the string is not actually safe, you have introduced an XSS vulnerability. Consequently, you should avoid calling this if possible. You can use @{function@libphutil:phutil_escape_html_newlines} to escape HTML while converting newlines to `
`. You should not need to explicitly use @{function@libphutil:phutil_escape_html} anywhere. If you need to apply a string function (such as `trim()`) to safe HTML, use @{method@libphutil:PhutilSafeHTML::applyFunction}. If you need to extract the content of a @{class@libphutil:PhutilSafeHTML} object, you should call `getHTMLContent()`, not cast it to a string. Eventually, we would like to remove the string cast entirely. Functions @{function@libphutil:phutil_tag} and @{function@libphutil:hsprintf} are not safe if you pass the user input for the tag or attribute name. All the following examples are dangerous: counterexample phutil_tag($evil); phutil_tag('span', array($evil => $evil2)); phutil_tag('span', array('onmouseover' => $evil)); // Use PhutilURI to check if $evil is valid HTTP link. hsprintf('', $evil); hsprintf('<%s>%s', $evil, $evil2, $evil); // We have a lint rule disallowing this. hsprintf($evil);