Learn R Programming

mvbutils (version 2.8.232)

maintain.packages: Set up task package for live editing

Description

See mvbutils.packaging.tools before reading or experimenting!

Set up task package(s) for editing and/or live-editing. Usually called in .First or .First.task. You need to be cded into the parent task of your task-package. maintain.packages must be called before loading the package via library or require. The converse, unmaintain.package, is rarely needed; it's really only meant for when unpackage doesn't work properly, and you want a "clean slate" task package.

Usage

# E.g. in your .First, after library( mvbutils), or in...
# ... a '.First.task' above yr task-package
maintain.packages(..., character.only = FALSE, autopatch=FALSE)
unmaintain.package( pkg, character.only = FALSE)

Arguments

...

names of packages, unquoted unless character.only is TRUE. Package names must correspond to subtasks of the current task.

character.only

see above

pkg

name of package, unquoted unless character.only is TRUE.

autopatch

whether to patch.install out-of-date installed packages (default FALSE, but TRUE is common).

Maintained packages as tasks

If you use mvbutils to pre-build your package, then your package must exist as a task in the cd hierarchy. Older versions of mvbutils allowed you to cd to a maintained package, but this is now forbidden because of the scope for confusion. Thanks to maintain.packages, there is no compelling need to have the package/task at the top of the search path; fixr, move, etc work just fine without. If you really do want to cd to a maintained package, you must call unmaintain.package first.

One piece of cleanup that I recommend, is to move any subtasks of "mypack" one level up in the task hierarchy, and to remove the tasks object from "Splendid" itself, e.g. via something like:

  cd( task.above.splendid)
  tasks <- c( tasks, combined.file.paths( tasks[ "Splendid"], ..Splendid$tasks))
  # ... combined.file.paths is an imaginary function. Watch out if you've used relative paths!
  rm.pkg( tasks, pkg="Splendid")

Details

maintain.packages( mypack) loads a copy of your task-package "mypack" (as stored in its ".RData" file) into a environment ..mypack (an "in-memory-task-package"), which itself lives in the "mvb.session.info" environment on the search path. You don't normally need to know this, because normally you'd modify/create/delete objects in the package via fixr or fixr(..., pkg="mypack") or rm.pkg( ..., pkg="mypack"). But to move objects between the package and other tasks, you do need to refer to the in-memory task package, e.g. via move( ..., from=..Splendid, to=subtask/of/current). In most cases, you will be prompted afterwards for whether to save the task package on disk, but you can always do yourself via Save.pos( ..Splendid). Note that only these updates and saves only update the task package and the loaded package. To update the source package using the task package, call pre.install; to update the installed package on disk as well as the source package, call patch.install.

Creating new things

It's always safe to create new objects of any type in .GlobalEnv, then use move(newthing,.,..mypack). For a new function, you can shortcut this two-step process and create it directly in the in-memory maintained package, via fixr(..mypack$newfun); fixr will take care of synchronization with the loaded package. This also ought to work for text objects created via fixtext. Otherwise, use the two-step route, unless you have a good reason to do the following...

Directly modifying the maintained package

Rarely, you may have a really good reason to directly modify the contents of ..mypack, e.g. via

  ..mypack$newfun <<- function( x) whatever

You can do it, but there are two problems to be aware of. The first is that changes won't be directly propagated to the loaded package, possibly not even after patch.install (though they will be honoured when you library() the package again). That is definitely the case for general data objects, and I'm not sure about functions; however, successful propagation after patch.install may happen for a special objects such as mypack.DESCRIPTION and documentation objects. Hence my general advice is to use fixr or move.

The second, minor, problem is that you will probably forget to use <<- and will use <- instead, so that a local copy of ..mypack will be created in the current task. This is no big deal, and you can just rm the local copy; the local copy and the master copy in "mvb.session.info" both point to the same thing, and modifying one implies modifying the other, so that deleting the local copy won't lose your changes. Save detects accidental local copies of task packages, and omits them from the disk image, so there shouldn't be any problems next time you start R even if you completely forget about local/master copies.

Autopatch

If autopatch==TRUE, then maintain.packages will check whether the corresponding installed packages are older than the ".RData" files of the task packages. If they are, it will do a full patch.install; if not, it will still call patch.install but only to reverse-update any bundled DLLs (see pre.install), not to re-install the R-source. I find autopatch useful with packages containing C code, where a crash in the C code can cause R to die before the most recent R-code changes have been "committed" with patch.install. When you next start R, a call to maintain.packages with autopatch=TRUE will "commit" the changes before the package is loaded, because you have to call maintain.packages before library; this seems to be more reliable than running patch.install manually after library after a restart.

See Also

mvbutils.packaging.tools, fixr, pre.install, patch.installed, unpackage

Examples

Run this code
# NOT RUN {
# In your .First:
library( mvbutils)
maintain.packages( myfirstpack, mysecondpack, mythirdpack)
# or...
live.edit.list <- c( 'myfirstpack', 'mysecondpack', 'mythirdpack')
maintain.packages( live.edit.list, character.only=TRUE)
library( myfirstpack) # etc
# }

Run the code above in your browser using DataLab