This documentation is intended to assist contributors to pgBackRest by outlining some basic steps and guidelines for contributing to the project.
Code fixes or new features can be submitted via pull requests. Ideas for new features and improvements to existing functionality or documentation can be [submitted as issues](https://github.com/pgbackrest/pgbackrest/issues). You may want to check the [Project Boards](https://github.com/pgbackrest/pgbackrest/projects) to see if your suggestion has already been submitted.
Bug reports should be [submitted as issues](https://github.com/pgbackrest/pgbackrest/issues). Please provide as much information as possible to aid in determining the cause of the problem.
You will always receive credit in the [release notes](http://www.pgbackrest.org/release.html) for your contributions.
Coding standards are defined in [CODING.md](https://github.com/pgbackrest/pgbackrest/blob/master/CODING.md) and some important coding details and an example are provided in the [Coding](#coding) section below. At a minimum, unit tests must be written and run and the documentation generated before [submitting a Pull Request](#submitting-a-pull-request); see the [Testing](#testing) section below for details.
Some unit tests and all the integration tests require Docker. Running in containers allows us to simulate multiple hosts, test on different distributions and versions of PostgreSQL, and use sudo without affecting the host system.
If using a RHEL-based system, the CPAN XML parser is required to run `test.pl` and `doc.pl`. Instructions for installing Docker and the XML parser can be found in the `README.md` file of the pgBackRest [doc](https://github.com/pgbackrest/pgbackrest/blob/master/doc) directory in the section "The following is a sample RHEL/CentOS 7 configuration that can be used for building the documentation". NOTE that the "Install latex (for building PDF)" section is not required since testing of the docs need only be run for HTML output.
The following sections provide information on some important concepts needed for coding within pgBackRest.
### Memory Contexts
Memory is allocated inside contexts and can be long lasting (for objects) or temporary (for functions). In general, use `MEM_CONTEXT_NEW_BEGIN("SomeName")` for objects and `MEM_CONTEXT_TEMP_BEGIN()` for functions. See [memContext.h](https://github.com/pgbackrest/pgbackrest/blob/master/src/common/memContext.h) for more details and the [Coding Example](#coding-example) below.
### Logging
Logging is used for debugging with the built-in macros `FUNCTION_LOG_*()` and `FUNCTION_TEST_*()` which are used to trace parameters passed to/returned from functions. `FUNCTION_LOG_*()` macros are used for production logging whereas `FUNCTION_TEST_*()` macros will be compiled out of production code. For functions where no parameter is valuable enough to justify the cost of debugging in production, use `FUNCTION_TEST_BEGIN()/FUNCTION_TEST_END()`, else use `FUNCTION_LOG_BEGIN(someLogLevel)/FUNCTION_LOG_END()`. See [debug.h](https://github.com/pgbackrest/pgbackrest/blob/master/src/common/debug.h) for more details and the [Coding Example](#coding-example) below.
Logging is also used for providing information to the user via the `LOG_*()` macros, such as `LOG_INFO("some informational message")` and `LOG_WARN_FMT("no prior backup exists, %s backup has been changed to full", strZ(cfgOptionDisplay(cfgOptType)))` and also via `THROW_*()` macros for throwing an error. See [log.h](https://github.com/pgbackrest/pgbackrest/blob/master/src/common/log.h) and [error.h](https://github.com/pgbackrest/pgbackrest/blob/master/src/common/error.h) for more details and the [Coding Example](#coding-example) below.
### Coding Example
The example below is not structured like an actual implementation and is intended only to provide an understanding of some of the more common coding practices. The comments in the example are only here to explain the example and are not representative of the coding standards. Refer to the Coding Standards document ([CODING.md](https://github.com/pgbackrest/pgbackrest/blob/master/CODING.md)) and sections above for an introduction to the concepts provided here. For an actual implementation, see [db.h](https://github.com/pgbackrest/pgbackrest/blob/master/src/db/db.h) and [db.c](https://github.com/pgbackrest/pgbackrest/blob/master/src/db/db.c).
#### Example: hypothetical basic object construction
```c
/*
* HEADER FILE - see db.h for a complete implementation example
*/
// Typedef the object declared in the C file
typedef struct MyObj MyObj;
// Constructor, and any functions in the header file, are all declared on one line
MyObj *myObjNew(unsigned int myData, const String *secretName);
// Declare the publicly accessible variables in a structure with Pub appended to the name
typedef struct MyObjPub // First letter upper case
{
MemContext *memContext; // Pointer to memContext in which this object resides
unsigned int myData; // Contents of the myData variable
} MyObjPub;
// Declare getters and setters inline for the publicly visible variables
// Only setters require "Set" appended to the name
__attribute__((always_inline)) static inline unsigned int
myObjMyData(const MyObj *const this)
{
return THIS_PUB(MyObj)->myData; // Use the built-in THIS_PUB macro
A list of all possible test combinations can be viewed by running:
```
pgbackrest/test/test.pl --dry-run
```
While some files are automatically generated during `make`, others are generated by running the test harness as follows:
```
pgbackrest/test/test.pl --gen-only
```
Prior to any submission, the html version of the documentation should also be run and the output checked by viewing the generated html on the local file system under `pgbackrest/doc/output/html`. More details can be found in the pgBackRest [doc/README.md](https://github.com/pgbackrest/pgbackrest/blob/master/doc/README.md) file.
```
pgbackrest/doc/doc.pl --out=html
```
> **NOTE:** `ERROR: [028]` regarding cache is invalid is OK; it just means there have been changes and the documentation will be built from scratch. In this case, be patient as the build could take 20 minutes or more depending on your system.
-`--vm` - identifies the pre-built container when using Docker, otherwise the setting should be `none`. See [test.yml](https://github.com/pgbackrest/pgbackrest/blob/master.github/workflows/test.yml) for a list of valid vm codes noted by `param: test`.
Build a container to run tests. The vm must be pre-configured but a variety are available. A vagrant file is provided in the test directory as an example of running in a virtual environment. The vm names are all three character abbreviations, e.g. `u20` for Ubuntu 20.04.
- **define.yaml** - defines the number of tests to be run for each module and test file. There is a comment at the top of the file that provides more information about this file.
- **src/module/somefileTest.c** - where "somefile" is the path and name of the test file where the unit tests are located for the code being updated (e.g. `src/module/command/expireTest.c`).
Each module is separated by a line of asterisks (*) and each test within is separated by a line of dashes (-). In the example below, the module is `command` and the unit test is `check`. The number of calls to `testBegin()` in a unit test file will dictate the number following `total:`, in this case 4. Under `coverage:`, the list of files that will be tested.
Unit test files are organized in the `test/src/module` directory with the same directory structure as the source code being tested. For example, if new code is added to src/**command/expire**.c then test/src/module/**command/expire**Test.c will need to be updated.
Assuming that a test file already exists, new unit tests will either go in a new `testBegin()` section or be added to an existing section. Each such section is a test run. The comment string passed to `testBegin()` should reflect the function(s) being tested in the test run. Tests within a run should use `TEST_TITLE()` with a comment string describing the test.
The [harnessConfig.h](https://github.com/pgbackrest/pgbackrest/blob/master/test/src/common/harnessConfig.h) describes a list of functions that should be used when configuration options are required for a command being tested. Options are set in a `StringList` which must be defined and passed to the `HRN_CFG_LOAD()` macro with the command. For example, the following will set up a test to run `pgbackrest --repo-path=test/test-0/repo info` command on multiple repositories, one of which is encrypted:
Sometimes it is desirable to store or manipulate files before or during a test and then confirm the contents. The [harnessStorage.h](https://github.com/pgbackrest/pgbackrest/blob/master/test/src/common/harnessStorage.h) file contains macros (e.g. `HRN_STORAGE_PUT` and `TEST_STORAGE_GET`) for doing this. In addition, `HRN_INFO_PUT` is convenient for writing out info files (archive.info, backup.info, backup.manifest) since it will automatically add header and checksum information.
Tests are run and results confirmed via macros that are described in [harnessTest.h](https://github.com/pgbackrest/pgbackrest/blob/master/test/src/common/harnessTest.h). With the exception of TEST_ERROR, the third parameter is a short description of the test. Some of the more common macros are:
-`TEST_RESULT_VOID` - The function being tested returns a `void`. This is then usually followed by tests that ensure other actions occurred (e.g. a file was written to disk).
If a function being tested logs something with `LOG_WARN`, `LOG_INFO` or other `LOG_*()` macro, then the logged message must be cleared before the end of the test by using the `TEST_RESULT_LOG()/TEST_RESULT_LOG_FMT()` macros.
```
TEST_RESULT_LOG(
"P00 WARN: WAL segment '000000010000000100000001' was not pushed due to error [25] and was manually skipped: error");
Sometimes it is useful to use a child process for testing. Below is a simple example. See [harnessFork.h](https://github.com/pgbackrest/pgbackrest/blob/master/test/src/common/harnessFork.h) for more details.
A PostgreSQL libpq shim is provided to simulate interactions with PostgreSQL. Below is a simple example. See [harnessPq.h](https://github.com/pgbackrest/pgbackrest/blob/master/test/src/common/harnessPq.h) for more details.
Unit tests are run for all files that are listed in `define.yaml` and a coverage report generated for each file listed under the tag `coverage:`. Note that some files are listed in multiple `coverage:` sections for a module; in this case, each test for the file being modified should be specified for the module in which the file exists (e.g. `--module=storage --test=posix --test=gcs`, etc.) or, alternatively, simply run the module without the `--test` option. It is recommended that a `--vm` be specified since running the same test for multiple vms is unnecessary for coverage. The following example would run the test set from the **define.yaml** section detailed above.
> **NOTE:** Not all systems perform at the same speed, so if a test is timing out, try rerunning with another vm.
Because a test run has not been specified, a coverage report will be generated and written to the local file system under the pgBackRest directory `test/result/coverage/lcov/index.html` and a file with only the highlighted code that has not been covered will be written to `test/result/coverage/coverage.html`.
If 100 percent code coverage has not been achieved, an error message will be displayed, for example: `ERROR: [125]: c module command/check/check is not fully covered`
**Debugging with files**
Sometimes it is useful to look at files that were generated during the test. The default for running any test is that, at the start/end of the test, the test harness will clean up all files and directories created. To override this behavior, a single test run must be specified and the option `--no-cleanup` provided. Again, continuing with the check command, from **define.yaml** above, there are four tests. Below, test one will be run and nothing will be cleaned up so that the files and directories in `test/test-0` can be inspected.
The following is a small sample of a typical test output.
```
run 8 - expireTimeBasedBackup()
run 8/1 ------------- L2285 no current backups
000.002s L2298 empty backup.info
000.009s 000.007s L2300 no backups to expire
```
**run 8 - expireTimeBasedBackup()** - indicates the run number (8) within the module and the parameter provided to testBegin, e.g. `testBegin("expireTimeBasedBackup()")`
**run 8/1 ------------- L2285 no current backups** - this is the first test (1) in run 8 which is the `TEST_TITLE("no current backups");` at line number 2285.
**000.002s L2298 empty backup.info** - the first number, 000.002s, is the time in seconds that the test started from the beginning of the run. L2298 is the line number of the test and `empty backup.info` is the test comment.
**000.009s 000.007s L2300 no backups to expire** - again, 000.009s, is the time in seconds that the test started from the beginning of the run. The second number, 000.007s, is the run time of the **previous** test (i.e. `empty backup.info` test took 000.007 seconds to execute). L2300 is the line number of the test and `no backups to expire` is the test comment.
Options can be added to a command or multiple commands. Options can be configuration file only, command-line only or valid for both. Once an option is successfully added, the `config.auto.h` and `parse.auto.c` files will automatically be generated by the build system.
There are detailed comment blocks above each section that explain the rules for defining commands and options. Regarding options, there are two types: 1) command line only, and 2) configuration file. With the exception of secrets, all configuration file options can be passed on the command line. To configure an option for the configuration file, the `section:` key must be present.
The `option:` section is broken into sub-sections by a simple comment divider (e.g. `# Repository options`) under which the options are organized alphabetically by option name. To better explain this section, two hypothetical examples will be discussed. For more details, see [config.yaml](https://github.com/pgbackrest/pgbackrest/blob/master/src/build/config/config.yaml).
Note that `section:` is not present thereby making this a command-line only option defined as follows:
-`set` - the name of the option
-`type` - the type of the option. Valid values for types are: `boolean`, `hash`, `integer`, `list`, `path`, `size`, `string`, and `time`
-`command` - list each command for which the option is valid. If a command is not listed, then the option is not valid for the command and an error will be thrown if it is attempted to be used for that command. In this case the valid commands are `backup` and `restore`.
-`backup` - details the requirements for the `--set` option for the `backup` command. It is dependent on the option `--stanza`, meaning it is only allowed to be specified for the `backup` command if the `--stanza` option has been specified. And `required: false` indicates that the `--set` option is never required, even with the dependency.
-`restore` - details the requirements for the `--set` option for the `restore` command. Since `required:` is omitted, it is not required to be set by the user but it is required by the command and will default to `latest` if it has not been specified by the user.
-`command-role` - defines the processes for which the option is valid. `main` indicates the option will be used by the main process and not be passed on to other local/remote processes.
#### Example 2: hypothetical configuration file option
-`section` - the section of the configuration file where this option is valid (omitted for command line only options, see [Example 1](#example-1-hypothetical-command-line-only-option-) above)
-`group` - indicates that this option is part of the `repo` group of indexed options and therefore will follow the indexing rules e.g. `repo1-test-type`.
-`default` - sets a default for the option if the option is not provided when the command is run. The default can be global (as it is here) or it can be specified for a specific command in the command section (as in [Example 1](#example-1-hypothetical-command-line-only-option-) above).
-`command` - list each command for which the option is valid. If a command is not listed, then the option is not valid for the command and an error will be thrown if it is attempted to be used for that command. In this case the valid commands are `backup` and `restore`.
-`command-role` - defines the processes for which the option is valid. `main` indicates the option will be used by the main process and not be passed on to other local/remote processes.
At compile time, the `config.auto.h` file will be generated to contain the constants used for options in the code. For the C enums, any dashes in the option name will be removed, camel-cased and prefixed with `cfgOpt`, e.g. `repo-path` becomes `cfgOptRepoPath`.
### reference.xml
All options must be documented or the system will error during the build. To add an option, find the command section identified by `command id="COMMAND"` section where `COMMAND` is the name of the command (e.g. `expire`) or, if the option is used by more than one command and the definition for the option is the same for all of the commands, the `operation-general title="General Options"` section.
To add an option, add the following to the `<option-list>` section; if it does not exist, then wrap the following in `<option-list>``</option-list>`. This example uses the boolean option `force` of the `restore` command. Simply replace that with your new option and the appropriate `summary`, `text` and `example`.
<text>By itself this option forces the <postgres/> data and tablespace paths to be completely overwritten. In combination with <br-option>--delta</br-option> a timestamp/size delta will be performed instead of using checksums.</text>
To quickly view the HTML documentation, the `--no-exe` option can be passed to the documentation generator in order to bypass executing the code elements:
If Docker is installed, it will be used by the documentation generator to execute the code elements while building the documentation, therefore, the `--no-exe` should be omitted, (i.e. `pgbackrest/doc/doc.pl --output=html`). `--no-cache` may be used to force a full build even when no code elements have changed since the last build. `--pre` will reuse the container definitions from the prior build and saves time during development.
The containers created for documentation builds can be useful for manually testing or trying out new code or features. The following demonstrates building through just the `quickstart` section of the `user-guide` without encryption.
The resulting Docker containers can be listed with `docker ps` and the container can be entered with `docker exec doc-pg-primary bash`. Additionally, the `-u` option can be added for entering the container as a specific user (e.g. `postgres`).
- Does it meet the [coding standards](https://github.com/pgbackrest/pgbackrest/blob/master/CODING.md)?
- Have [Unit Tests](#writing-a-unit-test) been written and [run](#running-a-unit-test) with 100% coverage?
- If your submission includes changes to the help or online documentation, have the [help](#testing-the-help) and [documentation](#testing-the-documentation) tests been run?
- Has it passed continuous integration testing? Simply renaming your branch with the appendix `-cig` and pushing it to your GitHub account will initiate GitHub Actions to run CI tests.
When submitting a Pull Request:
- Provide a short submission title.
- Write a detailed comment to describe the purpose of your submission and any issue(s), if any, it is resolving; a link to the GitHub issue is also helpful.
After submitting a Pull Request:
- One or more reviewers will be assigned.
- Respond to any issues (conversations) in GitHub but do not resolve the conversation; the reviewer is responsible for ensuring the issue raised has been resolved and marking the conversation resolved. It is helpful to supply the commit in your reply if one was submitted to fix the issue.