fix old serialized S4 instances

Introduction

r Biocpkg("updateObject") is an R package that provides a set of tools built around the updateObject() generic function to make it easy to work with old serialized S4 instances.

The package is primarily useful to package maintainers who want to update the serialized S4 instances included in their package.

Out-of-sync objects

Out-of-sync objects (a.k.a. outdated or old objects) are R objects that got serialized at some point and became out-of-sync later on when the authors/maintainers of an S4 class made some changes to the internals of the class.

A typical example of this situation is when some slots of an S4 class A get added, removed, or renamed. When this happens, any object of class A (a.k.a. A instance) that got serialized before this change (i.e. written to disk with saveRDS(), save(), or serialize()) becomes out-of-sync with the new class definition.

Note that this is also the case of any A derivative (i.e. any object that belongs to a class that extends A), as well as any object that contains an A instance or derivative. For example, if B extends A, then any serialized list of A or B objects is now an old object, and any S4 object of class C that has A or B objects in some of its slots now is also an old object.

An important thing to keep in mind is that, in fact, the exact parts of a serialized object x that are out-of-sync with their class definition can be deeply nested inside x.

The updateObject() generic function

updateObject() is the core function used in Bioconductor for updating old R objects. The function is an S4 generic currently defined in the r Biocpkg("BiocGenerics") package and with dozens of methods defined across many Bioconductor packages. For example, the r Biocpkg("S4Vectors") package defines updateObject() methods for Vector, SimpleList, DataFrame, and Hits objects, the r Biocpkg("SummarizedExperiment") package defines methods for SummarizedExperiment, RangedSummarizedExperiment, and Assays objects, the r Biocpkg("MultiAssayExperiment") package defines a method for MultiAssayExperiment objects, the r Biocpkg("QFeatures") package a method for QFeatures objects, etc...

See ?BiocGenerics::updateObject in the r Biocpkg("BiocGenerics") package for more information.

A tedious process

Serialized objects are typically (but not exclusively) found in R packages. To update all the serialized objects contained in a given package, one usually needs to perform the following steps:

Identify all the files in the package that contain serialized R objects. Serialized R objects are normally written to RDS or RDA files. These files typically use file extensions .rds (for RDS files), and .rda or .RData (for RDA files).
Load each serialized object into R. This is usually done by calling readRDS() on each RDS file, and load() on each RDA file. Note that unlike RDS files which can only contain a single object per file, RDA files can contain an arbitrary number of objects per file.
Pass each object thru updateObject():

x <- updateObject(x)

Note that if x doesn't contain any out-of-sync parts then updateObject() will act as a no-op, that is, it will return an object that is strictly identical to the original object.

Write each object back to its original file. This is done with saveRDS() or save(), depending on whether the object came from an RDS or RDA file. Note that this only needs to be done for objects that actually contained out-of-sync parts i.e. for objects on which updateObject() did not act as a no-op.

In addition to the above steps, the package maintainer also needs to perform the usual steps required for updating a package and publishing its new version. In the case of a Bioconductor package, these steps are:

Bump the package version.
Set its Date field (if present) to the current date.
Commit the changes.
Push the changes to git.bioconductor.org.

Performing all the above steps manually can be tedious and error prone, especially if the package contains many serialized objects, or if the entire procedure needs to be performed on a big collection of packages. The r Biocpkg("updateObject") package provides a set of tools that intend to make this much easier.

updateBiocPackageRepoObjects()

updateBiocPackageRepoObjects() is the central function in the r Biocpkg("updateObject") package. It takes care of updating the serialized objects contained in a given Bioconductor package by performing all the steps described in the previous section.

Let's load r Biocpkg("updateObject"):

library(updateObject)

## Set fake git user to make git_commit() happy:
set_git_user_name("titi")
set_git_user_email("titi@gmail.com")

and try updateBiocPackageRepoObjects() on the RELEASE_3_13 branch of the r Biocpkg("TimiRGeN") package:

repopath <- file.path(tempdir(), "TimiRGeN")
updateBiocPackageRepoObjects(repopath, branch="RELEASE_3_13", use.https=TRUE)

Important notes:

By default updateBiocPackageRepoObjects() does not try to push the changes to git.bioconductor.org. Only the authorized maintainers of the r Biocpkg("TimiRGeN") package can do that. If you are using updateBiocPackageRepoObjects() on a package that you maintain and you wish to push the changes to git.bioconductor.org, then do NOT use HTTPS access (i.e. don't use use.https=TRUE) and use push=TRUE.
The RELEASE_3_13 branch of all Bioconductor packages got frozen in October 2021. The above example is for illustrative purpose only. A more realistic situation would be to use updateBiocPackageRepoObjects() on the development version (i.e. the master branch) of a package that you maintain, and to push the changes by calling the function with push=TRUE:

updateBiocPackageRepoObjects(repopath, push=TRUE)

See ?updateBiocPackageRepoObjects for more information and more examples.

List of tools provided by the updateObject package

The package provides the following tools:

updateBiocPackageRepoObjects(): See above.
updatePackageObjects(): A simpler version of updateBiocPackageRepoObjects() that doesn't know anything about Git. That is, updatePackageObjects() will do the same thing as updateBiocPackageRepoObjects() except that it won't commit or push the changes. This means that the function can be used on any local package source tree, whether it's a Git clone or not, and whether it's a Bioconductor package or not.
updateAllBiocPackageRepoObjects() and updateAllPackageObjects(): Similar to updateBiocPackageRepoObjects() and updatePackageObjects() but for processing a set of Bioconductor package Git repositories (for updateAllBiocPackageRepoObjects()) and a set of packages (for updateAllPackageObjects()).
updateSerializedObjects(): The workhorse behind the above functions.