This is the source code for the "survival" package in R. It gets posted to the comprehensive R archive (CRAN) at intervals, each such posting preceded a throrough test. (I run the test suite for all 600+ packages that depend on survival.) In general, each new push to CRAN will update the second term of the version number, e.g. 2.40-5 to 2.41-0. Updates only to the github source increment after the dash. (If an error is found in the process of CRAN submission then the published CRAN version may be x.yy-1 or even x.yy-2 or 3.)
The vignette2 directory contains material that is not posted to CRAN. The file "tutorial.Rnw", for instance, requires data from the mstate package, survival is a recommended package, and such packages can only depend on other recommended packages. (This allows for a consistent distribution bundle.) The sas.Rnw vignette has a discussion of compute time and takes too long to run, etc.
A large portion of the source is found in the noweb directory, and is based on the literate programming ideas of Knuth. The reason is that it allows more complete documentation of the methods. I can have things like blocks of equations, and find having the "real" equations side by side with the code makes it much easier to get it right. Anyone who wants to study the methods is advised to perform "make code.pdf" in the noweb directory and then look at the relevant portion of that pdf file. Any file in the R or src directories that starts with an "automatically generated ..." comment should NOT be modified directly, instead work with the noweb source. (You will need to have the noweb package loaded in order to run the Makefile.)
You should be able to install this using the following R code:
Note that good practice would be to make derived files such as R/tmerge.R "on the fly" using a configure script; that way there would not be a danger of someone trying to modify the derived file rather than the actual source (noweb/tmerge.Rnw). However, I was not able to create a configure file that worked reliably on all platforms, and voted for usability rather than purity.