From Andrei Gaponenko, 11 December 2003.  (Tweaked by RPM)
https://twist.phys.ualberta.ca/forum/index.php?site=twist&bn=twist_software&action=view&key=1071196547

Some of the ROOT trees in the systematics/ directory cause analysis
jobs to crash, or it may be desirable to exclude a run from an
analysis for other reasons.  Till now people were renaming run
directories under systematics/.../analN/root from "runNNN" to
something else to prevent that runs from being analyzed. 

The problem with this approach that after you have run an analysis on
"all good runs" of genXX/analYY and got the histograms, the definition
of "good" runs may be changed by someone, and you will not be able to
easily look at e.g. log files from all runs you analyzed, or to
reproduce the histograms.  (Note that it's possible for a run to pass
through one analysis, but to crash a different one.)

The proposed solution is:

1) Do not change anything under root/
   I.e. no renaming/removing of the runNNN directories.

2) To tell known "good" runs from "bad", use the goodlinks/ and
   badlinks/ directories.  The directories located under analN/, at
   the same level as the root/ directory.  Initially everything goes
   to goodlinks/, when we decide to exclude a run from the analysis
   its link gets moved to badlinks/, and possibly renamed to indicate
   the reason.  This must be done as user e614.  For example:
   
   mv goodlinks/run1 badlinks/
   cd badlinks/
   mv run1 run1.missing-mce2

   The directories must contain links to the *absolute* pathnames
   of the root/runNNN dirs.  I.e. to create the initial setup do 
   
   cd goodlinks/
   ln -s /twist/tw00z/systematics/...../anal1/root/* .

   *NOT* ls -s ../root/* .

   If ln complains that the command line is too long, the solution is

   bash
   for i in /twist/tw00z/systematics/...../anal1/root/*; do ln -s $i .; done
   
3) To run an analysis, you first copy the good links to your
   private area (under your usual username, not e614)

   mkdir my-linkdir
   cp -p -r /twist/tw00z/systematics/.../anal1/goodlinks/* .

   (thus the absolute paths for the links are important),
   then run your program on my-linkdir/*/tree*.root
   
   Even if someone moves a link from "good" to "bad" at this point,
   you are not affected. More importantly, a month later you still
   have a record (the bunch of symlinks) of exactly what runs were
   used to make a histogram.


I have created goodlinks/ and badlinks/ for all existing systematics,
and moved known bad runs to badlinks/.  Some older directories with
different organization (e.g. all root files in a single subdir, or
directory names like rhoNNN, etaNNN instead of runNNN) were not
handled properly, if someone wants to analyze these old data please
fix it yourself.