Tools

Before you start writing a new tool please search the Main Tool Shed (MTS) and the Test Tool Shed (TTS) because it’s possible that someone has already created a wrapper for the same third party executable you are looking for. Consider announcing your tool project on galaxy-dev to see if anyone has already created a wrapper.

Tool versions

Tool versions are mandatory to enable reproducibility. Version is an attribute of the XML tool element, e.g.

<tool id="rgTF" name="Tool Factory" version="1.11">

and should be incremented with each change of the wrapper that is released to the Tool Shed (except for cosmetic modifications).

The value should follow the PEP 440 specification.

If the Galaxy tool is a wrapper for an underlying tool, we recommend to:

  • define a @TOOL_VERSION@ macro token, which you should also re-use in the corresponding <requirement> element;
  • optionally, define a @VERSION_SUFFIX@ macro token, which may be placed either in the tool wrapper or in a shared macro file. This should be set to an integer number:
    • starting at 0 for the first wrapper release of each version of the underlying tool;
    • to be increased by 1 whenever you update the wrapper without changing the underlying @TOOL_VERSION@.
  • set the tool version attribute to @TOOL_VERSION@+galaxy@VERSION_SUFFIX@. If it is preferred not to use a @VERSION_SUFFIX@ token (e.g. to allow bumping the version only for a specific tool in a suite), the tool version attribute should be simply set to @TOOL_VERSION@+galaxyN, where N is an integer following the same rules as @VERSION_SUFFIX@.

If instead the Galaxy tool cannot be identified with a single underlying tool, the +galaxy@VERSION_SUFFIX@ local version identifier should be omitted, and any version value can be used, as long as it respects the PEP 440 specification.

For tools whose wrapper version is (for historical reasons) already greater than the version of the underlying tool, only the minor version number shall be increased if this is likely to bring the two version in sync in a reasonable time.

Tool ids

Should be meaningful and unique also in a larger context. If your tool is called grep try to prefix that name with something meaningful. Objective is to make it easier for Galaxy admins to identify a tool based on the short ID. Otherwise they would need to use the long toolshed/xx/ id.

Some simple rules for generating tool IDs:

  • Tool IDs should contain only [a-z0-9_-].
  • Multiple words should be separated by underscore or dashes
  • Suite tools should prefix their ids with the suite name. E.g. bedtools_*

Tool Names

Names are important! Names are how users and admins find your tools. Names should strive to be unique within a suite of tools, and may wish to include the suite name if it is a well known suite. Some instructional examples:

  • Cufflinks, Cuffdiff, Cuffmerge are in a suite together.
  • The vsearch suite contains tools with names like “VSearch Alignment”, “VSearch Clustering”, etc.

In the cufflinks example, everyone knows the functionality of the cufflinks command, and can easily guess as the use of a tool named “cuffdiff” in their tool panel.

With VSearch however, a tool named “Alignment” would not be useful, as users would have a hard time finding it and gathering context about its functionality. With the VSearch prefix, once a user learns what one VSearch tool does, they can quickly apply that to the other available VSearch tools.

Tool Descriptions

Tool names are not your only tool for making your tool discoverable to end users, and conveying information regarding the functionality of said tool. Tool descriptions are displayed directly after the tool name and generally conform to a “sentence” like structure.

  • bowtie2 is a short read aligner
  • Cuffmerge merges together several Cufflinks assemblies
  • NCBI BLAST+ database info shows BLAST database information from blastdbcmd

In the above examples the tool name is rendered in fixed width text, and the rest is the tool description.

Parameter name, argument and help

The argument attribute of <param> should include the long form of the underlying tool parameter, e.g. argument="--max". This is automatically displayed inside the parameter help and is useful to give the user the chance to go to the original documentation and map the Galaxy UI element to the actual parameter. It also makes debugging easier if the user is talking to non-Galaxy developers.

When argument is specified, the name attribute becomes optional and, if not included, is derived from argument by stripping any leading dashes and replacing internal dashes by underscores (the later since release 19.09). This derived name can be used inside the <command> element to refer to the parameter value as you would normally do with the name attribute. Note that if the automatically generated name violates the rules for valid Cheetah placeholders (i.e. consist of alphanumeric characters or underscore and must not start with a digit) you should specify a valid name attribute for the parameter.

Tests

All Galaxy Tools should include functional tests. In their simplest form, you provide sample input files and expected output files for given parameter values. Where the output file is not entirely reproducible you can make assertions about the output file contents.

Testing error conditions is also important. Recent development now allows tests say if the test should fail, and to make assertions about the tool’s stdout and stderr text (e.g. check expected summary text or warning messages appear). See planemo docs for more information.

When tools contain output filters, tests should be included that verify this filtering occurs. See planemo docs for more information.

Booleans

truevalue and falsevalue attributes of <param> should contain the underlying tool parameter. This makes it really easy to reference the param name in the Cheetah <command> section.

<command>
...
$strict
...
</command>
<inputs>
    ...
    <param name=”strict” truevalue=”--enable-strict” falsevalue=””>

Boolean should not be used as a conditional for other options. For dynamic options, please use a select input type as described in the Dynamic Options section below.

Dynamic Options

Options that are conditionally hidden (using the <conditional> element) should use a select param type and not a boolean. The user may not expect a boolean checkbox to change the content of a form.

To create an “Advanced options” section which is normally hidden and the user can expand, a <section> element can be used instead of a <conditional>. Beware that parameters inside a hidden section still have a value set, which is used when creating the job command, while in a “closed” conditional the non-visible parameters don’t have a value.

Command tag

The command tag is one of the most important parts of the tool, next to the user-facing options. It should be highly legible.

Command Formatting

The command tag should be started and finished by a CDATA tag, allowing direct use of characters like the ampersand (&) without needing XML escaping (&amp;).

<![CDATA[ your lines of Cheetah here ]]>

Wikipedia has more on CDATA

All Cheetah variables for text parameters, input and output files must be single-quoted, e.g. '${var_name}'.

For composite datatypes the recommended attribute to access the associated directory name differs for inputs (e.g. $input.extra_files_path) versus outputs (e.g. $output.files_path). This difference is historical, and it is hoped this will be harmonised in a future Galaxy release.

If you need to execute more than one shell command, concatenate them with a double ampersand (&&), so that an error in a command will abort the execution of the following ones.

Exit Code Detection

Unless the tool has special requirements, you should take advantage of the exit code detection provided by Galaxy, in lieu of using the <stdio/> tags. This can be done by adding a detect_errors tag to your <command /> block like so:

<command detect_errors="aggressive">
...
</command>

This will automatically fail the tool if the exit code is non-zero, or if the phrases error: or exception: appear in STDERR.

Help tag

The help tag should be started and finished by a CDATA tag.

<![CDATA[ your lines of restructuredText here ]]>

http://en.wikipedia.org/wiki/CDATA

Inside the help tag you should describe the functionality of your tool. The help tag is to the help="" attribute as a man page is to the --help flag. The help tag should cover the tools functionality, use cases, and even known issues in detail. The help tag is a good place to provide examples of how to run the tool and discuss specific subcases that your users might be interested in.

Including Images

If you have produced images detailing how your tool works (e.g. bedtools), it might be nice for those images to be included in the Galaxy tool documentation!

Images should be placed in a subdirectory, ./static/images/, and referenced in your tool help as .. image:: my-picture.png. This can be seen in the IUC’s wrappers, such as the one for the bedtools slop command.

Tool Dependency Package

If you are using perl/ruby/python/R packages, use the corresponding *_environment tags to depend on a specific version of Perl/Ruby …

Generating Indices

Occasionally data needs to be indexed (e.g. bam, fasta) files. When data is indexed, those indices should be generated in the current working directory rather than alongside the input dataset. This is part of the tool contract, you can read from your inputs, but only write to your outputs and CWD.

It’s convenient to do something like:

ln -sfn "${input_fasta}" tmp.fa;

before data processing in order to be able to easily generate the indices without attempting to write to a (possibly) read-only data source.

Datatypes

For now, the recommended practice is to push new datatypes to the Galaxy repository.

Data Managers

TODO

Coding Style