This document outlines the structure of CodeQL query files. You should adopt this structure when contributing custom queries to this repository, in order to ensure that new queries are consistent with the standard CodeQL queries.
* Metadata area–displayed at the top of the file, contains the metadata that defines how results for the query are interpreted and gives a brief description of the purpose of the query.
* Query definition–defined using QL. The query includes a select statement, which defines the content and format of the results. For further information about writing QL, see the following topics:
Query file metadata contains important information that defines the identifier and purpose of the query. The metadata is included as the content of a valid [QLDoc](https://codeql.github.com/docs/ql-language-reference/ql-language-specification/#qldoc) comment, on lines with leading whitespace followed by `*`, between an initial `/**` and a trailing `*/`. For example:
To help others use your query, you should include all of the required information outlined below in the metadata, and as much of the optional information as possible. For further information on query metadata see [Metadata for CodeQL queries](https://codeql.github.com/docs/writing-codeql-queries/metadata-for-codeql-queries/) on codeql.github.com.
You must specify an `@name` property for your query. This property defines the display name for the query. Query names should use sentence capitalization, but not include a full stop. For example:
You must define an `@description` property for your query. This property defines a short help message. Query descriptions should be written as a sentence or short-paragraph of plain prose, with sentence capitalization and full stop. The preferred pattern for alert queries is "Syntax X causes behavior Y." Any code elements included in the description should be enclosed in single quotes. For example:
You must specify an `@id` property for your query. It must be unique and should follow the standard CodeQL convention. That is, it should begin with the 'language code' for the language that the query analyzes followed by a forward slash. The following language codes are supported:
Further terms can be added to the `@id` to group queries that, for example, highlight similar issues or are of particular relevance to a certain framework. For example:
Note, `@id` properties should be consistent for queries that highlight the same issue for different languages. For example, the following queries identify format strings that contain unsanitized input in Java and C++ code respectively:
Alert queries (`@kind problem` or `path-problem`) support two further properties. These are added by GitHub staff after the query has been tested. The following information is for reference:
*`@security-severity`-defines the level of severity, between 0.0 and 10.0, for queries with `@tags security`. For more information about calculating `@security-severity`, see the [GitHub changelog](https://github.blog/changelog/2021-07-19-codeql-code-scanning-new-severity-levels-for-security-alerts/).
The `@tags` property is used to define categories that the query relates to. Each alert query should belong to one (or more, if necessary) of the following four top-level categories:
Metric queries (`@kind metric`) may have the `summary` tag. If SARIF output is used, the results of these queries can be found at `run[].properties.metricResults`.
If necessary, you can also define your own low-level tags to categorize the queries specific to your project or organization. When creating your own tags, you should:
* Use all lower-case letters, including for acronyms and proper nouns, with no spaces. All characters apart from * and @ are accepted.
* Use a forward slash / to indicate a hierarchical relationship between tags if necessary. For example, a query with tag `foo/bar` is also interpreted as also having tag `foo`, but not `bar`.
* Use a single-word `@tags` name. Multiple words, separated with hyphens, can be used for clarity if necessary.
If your query is a security query, use one or more `@tags` to associate it with the relevant CWEs. Add `@tags` for the most specific Base Weakness or Class Weakness in [View 1000](https://cwe.mitre.org/data/definitions/1000.html), using the parent/child relationship. For example:
When you tag a query like this, the associated CWE pages from [MITRE.org](https://cwe.mitre.org/index.html) will automatically appear in the reference section of its associated qhelp file.
Code Scanning may use tags to identify queries with specific meanings across languages. Currently, there is only one such tag: `lines-of-code`. The sum of the results for queries with this tag that return a single number column ([example for JavaScript](https://github.com/github/codeql/blob/c47d680d65f09a851e41d4edad58ffa7486b5431/java/ql/src/Metrics/Summaries/LinesOfCode.ql)) is interpreted by Code Scanning as the lines of code under the source root present in the database. Each language should have exactly one query of this form.
Maintainers are expected to add a `@security-severity` tag to security relevant queries that will be run on Code Scanning. There is a documented internal process for generating these `@security-severity` values.
The select clause of each alert query defines the alert message that is displayed for each result found by the query. Alert messages are strings that concisely describe the problem that the alert is highlighting and, if possible, also provide some context. For consistency, alert messages should adhere to the following guidelines:
* Each message should be a complete, standalone sentence. That is, it should be capitalized and have proper punctuation, including a full stop.
* The message should factually describe the problem that is being highlighted–it should not contain recommendations about how to fix the problem or value judgements.
* Program element references should be in 'single quotes' to distinguish them from ordinary words. Quotes are not needed around substitutions (`$@`).
* Avoid constant alert message strings and include some context, if possible. For example, `The class 'Foo' is duplicated as 'Bar'.` is preferable to `This class is duplicated here.`
* If a reference to the current location can't be avoided use "this location" instead of "here". For example, `Bad thing at this location.` is preferable to `Bad thing here.`. This avoids the "click here" anti-pattern.
* For path queries, if possible, try to follow the template: `This path depends on a [user-provided value].`, or alternatively (if the first option doesn't work) `[User-provided value] flows to this location and is used in a path.`.
* Where you reference another program element, link to it if possible using a substitution (`$@`). Links should be used inline in the sentence, rather than as parenthesised lists or appositions.
* Avoid using link texts that don't describe what they link to. For example, rewrite `This sensitive data is written to a logfile unescaped [here]` to `This sensitive data is [written to a logfile unescaped]`.
* Make link text as concise and precise as possible. For example, avoid starting a link text with an indefinite article (a, an). `Path construction depends on a [user-provided value]` is preferable to `Path construction depends on [a user-provided value]`. (Where the square brackets indicate a link.) See [the W3C guide on link texts](https://www.w3.org/WAI/WCAG22/Understanding/link-purpose-in-context.html) for further information.
* When a message contains multiple links, construct a sentence that has the most variable link (that is, the link with most targets) last. For further information, see [Defining the results of a query](https://codeql.github.com/docs/writing-codeql-queries/defining-the-results-of-a-query/).
For further information on query writing, see [CodeQL queries](https://codeql.github.com/docs/writing-codeql-queries/codeql-queries/). For more information on learning CodeQL, see [CodeQL documentation](https://codeql.github.com/docs/).
- This indicates a metric with a specific location in the codebase, for example the lines of code within a file. The `entity` here must have a valid location in the source code.
