Development practices

Making changes to GDAL

Minor changes to GDAL, such as bug fixes, may be made by opening a GitHub pull request.

Major changes should be discussed on the gdal-dev listserv and may require the drafting of a RFC (request for comment) document.

GDAL's policy on substantial code additions is documented at RFC 85: Policy regarding substantial code additions.


GDAL strives to be widely portable to 32 bit and 64 bit computing environments, as well as little-endian and big-endian ordered CPUs. CPL functions in the port directory provide services to abstract platform specific operations.

Generally speaking, where available CPL functions should be used in preference to operating system functions for operations like memory allocation, path parsing, filesystem I/O (using VSILFILE* / VSIVirtualFile*), ODBC access, etc.

C/C++ standards

The current C and C++ standards adopted by GDAL/OGR are C99 and C++17 (last updated per RFC 98: Build requirements for GDAL 3.9).

Variable naming

Much of the existing GDAL/OGR code uses an adapted Hungarian naming convention. Use of this convention is not mandatory, but when maintaining code using this convention it is desirable to continue adhering to it with changes. Most importantly, please avoiding using it improperly as that can be very confusing.

In Hungarian prefixing the prefix tells something about about the type, and potentially semantics of a variable. The following are some prefixes used in GDAL/OGR.

  • a: array

  • b: C/C++ bool. In C code that pre-dates C99 adoption, it is also used for ints with only TRUE/FALSE values.

  • by: byte (GByte / unsigned char).

  • df: floating point value (double precision)

  • e: enumeration

  • i: integer number used as a zero based array or loop index.

  • f: floating point value (single precision)

  • h: an opaque handle (such as GDALDatasetH).

  • n: integer number (size unspecified)

  • o: C++ object

  • os: CPLString or std::string

  • p: pointer

  • psz: pointer to a null-terminated string. (eg. "char *pszName;")

  • sz: null-terminated string (eg." char szName[100];")

  • k: compile-time constant

Prefixes can be stacked. The following are some examples of meaningful variables.

  • *char !*papszTokens: Pointer to an array of strings.

  • *int panBands: Pointer to the first element of an array of numbers.

  • *double padfScanline: Pointer to the first element of an array of doubles.

  • *double pdfMeanRet: Pointer to a single double.

  • *GDALRasterBand poBand: Pointer to a single object.

  • *GByte pabyHeader: Pointer to an array of bytes.

It may also be noted that the standard convention for variable names is to capitalize each word in a variable name.

Function and class naming

  • Functions and classes should have a selective enough namespace ("GDAL" or "OGR" prefix, or use of C++ namespace) to avoid symbol collision.

File naming and code formatting

  • All source files (.h, .c, .cpp, .py, etc.) should have a header with copyright attribution and the text of the GDAL X/MIT license.

  • Use lower case filenames.

  • Use .cpp extension for C++ files (not .cc).

  • C/C++ code formatting rules are defined in .clang-format. Python code formatting is enforced by Black. The pre-commit utility should be used to enforce them automatically. See Commit hooks.

Memory allocation

Large memory allocations should be performed using the VSIMalloc() family of functions, which will return nullptr on allocation failure. As per RFC 19: Safer memory allocation in GDAL, you can use VSIMalloc2(x, y) instead of doing CPLMalloc(x * y) or VSIMalloc(x * y). VSIMalloc2() will detect potential overflows in the multiplication and return a NULL pointer if it happens. This can be useful in GDAL raster drivers where x and y are related to the raster dimensions or raster block sizes. Similarly, VSIMalloc3(x, y, z) can be used as a replacement for CPLMalloc(x * y * z).

When working with standard library data structures such as std::vector that may throw std::bad_alloc, a try/catch block should be used around blocks that may allocate a large amount of memory.

Adding a new driver

  • If the driver depends on a third-party library, compilation of the driver must be made conditional on the presence of the library. Drivers should try to re-use existing library dependencies as much as possible, e.g. Expat for SAX XML parsing.

  • For a vector driver, check that the Open() method of the driver (often delegated to a Open() method of the datasource) is selective enough (i.e. it will not accept data files that are not meant for the driver), and robust enough (it will not crash for small variations w.r.t content that it would recognize). Check that it can deal with unusual filenames. For a GDAL driver, similar checks, as well for the optional Identify() method.

  • A set of tests covering the driver should be added to the Python test suite. If appropriate, small sample data files may be added to autotest/gdrivers/data or autotest/ogr/data. The test_ogrsf utility and GDALTest class may simplify testing of basic driver functionality.

  • A documentation page should be created for the driver. Documentation should, at a minimum, briefly describe the format handled by the driver and, when relevant, describe the particular syntax for the connection string, creation options, configuration options, etc. The documentation should provide a link to a more detailed format description and mention needed third-party libraries.

Writing tests

See Recommendations on how to write new tests.

Git usage

This section collects a few best practices for git usage for GDAL development.

Initiating your work repository

Fork OSGeo/gdal from the GitHub UI, and then run:

git clone
cd gdal
git remote add my_user_name

Working with a feature branch

git checkout master
# potentially update your local master against upstream, as described above
git checkout -b my_new_feature_branch

# do work. For example:
git add my_new_file
git add my_modifid_message
git rm old_file
git commit -a

# you may need to resynchronize against master if you need some bugfix
# or new capability that has been added since you created your branch
git fetch origin
git rebase origin/master

# At end of your work, make sure history is reasonable by folding non
# significant commits into a consistent set
git rebase -i master
# use 'fixup' for example to merge several commits together,
# and 'reword' to modify commit messages

# or alternatively, in case there is a big number of commits and marking
# all them as 'fixup' is tedious
git fetch origin
git rebase origin/master
git reset --soft origin/master
git commit -a -m "Put here the synthetic commit message"

# push your branch
git push my_user_name my_new_feature_branch

From the GitHub UI, issue a pull request.

If the pull request discussion or automated checks require changes, commit locally and push. To get a reasonable history, you may need to combine commits using git rebase -i master, in which case you will have to force-push your branch with git push -f my_user_name my_new_feature_branch.

Updating your local master against upstream master

git checkout master
git fetch origin

# Be careful: this will lose all local changes you might have done now
git reset --hard origin/master

Commit messages

Commit messages should indicate a component name (eg a driver name), a short description, and when relevant, a reference to a issue (with 'fixes #' if it actually fixes it)

COMPONENT_NAME: fix bla bla (fixes #1234)

Details here...

Commit hooks

GDAL provides pre-commit hooks to run code formatters and linters before a commit is made. The hooks are cloned with the repository and can be installed using pre-commit:

python3 -m pip install pre-commit
pre-commit install

Once installed, the hooks can be run manually via pre-commit run --all-files.

Blame ignore file

Due to whole-tree code reformatting done during GDAL 3.7 development, git blame information might be misleading. To avoid that, you need to modify your git configuration as following to ignore the revision of the whole-tree reformatting:

git config blame.ignoreRevsFile .git-blame-ignore-revs

Backporting bugfixes from master to a stable branch

git checkout master
With git log, identify the sha1sum of the commit you want to backport
git checkout 2.2 # if you want to backport to 2.2
git pull origin 2.2
# git checkout -b branch_name  # if you intend to submit the backport as a pull request
git cherry-pick the_sha1_sum
git push ...

If changes are needed, do them and git commit -a --amend

Things you should NOT do

(For anyone with push rights to OSGeo/gdal) Never modify a commit or the history of anything that has been committed to

Committing symbolic links is allowed only under the .github directory in order to avoid potential problems on Windows.

Source tree layout

  • alg: Algorithms: rasterization, polygonization, warper engine, etc.

  • apps: C++ command line utilities

  • autotest: Regression test suite (C++ and Python)

  • cmake: CMake modules and helpers functions

  • doc: Source code of GDAL documentation and scripts

  • docker: Dockerfile's for GDAL Docker images

  • gcore: Raster core functionality. Base classes: GDALDataset, GDALRasterBand, GDALDriver, overview building, etc.

  • frmts: GDAL/raster drivers (mostly, with the exception of the GDAL GeoPackage raster support in ogr/ogrsf_frmts/gpkg)

  • fuzzer: Source code and scripts for GDAL OSS-Fuzz integration

  • gnm: Source code for Geographic Networks Data Model

  • ogr: OGR vector core classes: OGRFieldDefn, OGRGeomFieldDefn, OGRFeatureDefn, OGRGeometry and derived classes, OGR SQL, etc.

  • ogr/ogrsf_frmts: OGR/vector drivers

  • ogr/ogrsf_frmts/generic: OGR vector core class: OGRLayer, OGR SQL generic layer

  • port: CPL (Common Portability Library)

  • perftests: C++ and Python scripts to check the speed/performance of various aspects of GDAL.

  • scripts: various utility scripts used for Continuous Integration, release generation, and other auxiliary tasks. None of them are for end users.

  • swig/include: definition of SWIG Python, Java, C# bindings

  • swig/python/gdal-utils/scripts: Launcher scripts for installed/public GDAL Python utilities. No real functionality there

  • swig/python/gdal-utils/osgeo_utils: Core code for GDAL Python utilities. Available in the PyPI gdal and gdal-utils packages.

  • swig/python/gdal-utils/samples: Scripts that are not installed and generally not or very little documented. May serve as a staging area for future scripts that are going to be promoted as official.

  • swig/python/gdal-utils/auxiliary: Helper methods and classes used by GDAL Python utilities. Available in the PyPI gdal and gdal-utils packages.

  • third_pary: Third-party libraries used by libgdal. Other may be found in:

    • alg/internal_libqhull

    • apps/argparse

    • frmts/gtiff/libtiff

    • frmts/gtiff/libgeotiff

    • frmts/hdf4/hdf-eos

    • frmts/jpeg/libjpeg

    • frmts/jpeg/libjpeg12

    • frmts/grib/degrib/degrib

    • frmts/grib/degrib/g2clib

    • frmts/pcidsk/sdk

    • frmts/pcraster/libcsf

    • frmts/png/libpng

    • frmts/gif/giflib

    • frmts/zlib/

    • ogr/ogrsf_frmts/cad/libopencad

    • ogr/ogrsf_frmts/geojson/libjson

    • ogr/ogrsf_frmts/flatgeobuf/flatbuffers

    • ogr/ogrsf_frmts/pmtiles/pmtiles

    • ogr/ogrsf_frmts/sqlite/sqlite_rtree_bulk_load