Addressing Bugs in nlme
Background
The nlme package (maintained by R-core) has a medium-sized list of bugs reported, reaching back over 5 years. In some cases there is already a patch or a proposed solution. Other cases require some discussion or further debugging to determine an appropriate solution. Often only the R code is affected and the problem is more about determining the correct behaviour rather than debugging low-level code, though that will be required in the hardest cases.
This project would suit a group of people working in twos or threes, with each small group aiming to close one or two bugs during the sprint.
The nlme package is version controlled in a Subversion repository, available at https://svn.r-project.org/R-packages/trunk/nlme/.
Problem statement
The list of nlme bugs on R’s Bugzilla as of 23 June 2023 is summarised below:
| Bug ID | Level | Status | Topic | Action(s) | Programming skills | Statistical knowledge |
|---|---|---|---|---|---|---|
| 18505 | Novice | patch available | behaviour changing with order of group variable | check patch | R | |
| 17884 | Novice | patch available | new confint.lme method for fixed effects only | check patch | R | confidence intervals |
| 17988 | Novice | suggested solution | what to return for corARMA with p = q = 0 | check patch, update doc | R | |
| 18472 | Intermediate | needs reprex | correct behaviour of gls | create reprex, discuss | R | gls |
| 17712 | Intermediate | suggested solution | conditions when algorithm should use decomposition | create patch, check patch | R | |
| 18285 | Intermediate | suggested solution | how to handle rank deficiency | discuss, create reprex, create patch | R | rank deficiency |
| 17961 | Intermediate | unclear if bug or not | appropriate df to use in tests | discuss | R | anova tests |
| 17866 | Intermediate | reprex available | Initializing corStruct outside gls call gives different results | debug | R | |
| 16890 | Intermediate | patch welcome | intervals.lmList does not work for with pool = FALSE | create patch | R | |
| 18192 | Intermediate | needs analysis | crash in nlme equivalent to lme model | debug R/C | C | |
| 17227 | Intermediate | partial analysis available | crash in nlme | debug R/C | C | |
| 16864 | Advanced | needs analysis | lme fails to warn when data are near-redundant | debug R/C/Fortran | R, C, Fortran | |
| 16975 | Advanced | needs analysis | REML estimation incorrect with fixed sigma | debug | R, C | |
| 17236 | Advanced | needs support from R Core | Correct df for intercept in lme | discuss, create patch, check patch | R | |
| 18433 | Advanced | needs reprex | infinite loop dependent on random seed | create reprex, debug | C, Docker | nlme algorithms |
| 16806 | Advanced | patch welcome | method fails for individuals with single observations | create patch | C |
Proposed solution
The table in the last section summarises the next action(s) required to address the different bugs. In more detail:
check patch
Check the bug is still present in the development version of nlme.
Apply the patch to the nlme sources, build and install the package, and test a reprex to confirm the patch fixes the issue.
Run R CMD check on the built package to check nothing is broken by the change.
update doc
Modify the source of the documentation file in the nlme sources.
Review your changes (e.g. Preview in RStudio or build and install to view).
Create a patch to include both code and documentation changes.
create reprex
- Create a minimal, reproducible example, using only base R packages plus nlme, that demonstrates the issue. In complex cases, this may require creating a Docker image so that the whole environment is reproducible.
discuss
- Discuss what the appropriate behaviour of the code should be, or discuss a proposal to clarify the way forward. Summarize conclusions on Bugzilla or agree a way forward with a member of R Core.
debug
- Identify the parts of the code that are the root cause of the issue. This may not always be code that triggers an error, but code that does not produce the correct result in particular circumstances.
create patch
- Modify the R and/or C source code to implement a proposed change. This can involve substantial work to determine the appropriate changes to make and may be an iterative process with feedback from others.
Testing
In general, we will rely on the tests included in the nlme package. It may be appropriate to propose additions to this test suite to cover the code added/modified when addressing a bug report.
We may also run reverse dependency checks to check changes do not break packages that depend on nlme.
Project requirements
Familiarity with Subversion will be helpful, however note that you can create a RStudio package project from the repository URL: https://svn.r-project.org/R-packages/trunk/nlme/.
Familiarity with R package development will be helpful, though again RStudio makes it simple to run R CMD check on the package or to build and install from source. Note that documentation in nlme is written directly in .Rd format.
On MacOS you should have the mandatory development tools, Xcode and GNU Fortran installed.
On Windows you should have RTools installed.
Project resources
Mixed-Effects Models in S and S-PLUS, José C. Pinheiro and Douglas M. Bates - we will have at least one copy of this book available at the sprint.
Project outcomes
Contribution to a recommended package (potentially acknowledged in the ChangeLog).
Experience with the package development process.
Experience with creating/applying patches with Subversion.
Experiences working with bug reports on Bugzilla.