RCS control of simulation projects in a single directory. The idea is to be able to reproduce a simulation given its version number. The version number was printed along with all the neuron windows on the screen when the simulation was archived. See Archive And Hardcopy
To manage parameter sets instead, use:
Effective use of this management system requires that the user conform to a policy in which no simulation variables are changed except by changing them in managed files (listed in the nrnversion file). ie. during a session only change variables by editing a hoc file and saving it, or only change variables in field editors that actually appear on the screen when a simulation is archived. Variable values will be lost if you change them by direct command to the interpreter (the command does not appear in any file) or if you dismiss a panel after changing one of its field editors (when the simulation is archived, only values of field editors actually displayed on the screen are saved). User judgment is required to determine if these missing variable values will later prevent reproduction of the simulation. If this is the case, then when the hardcopy is printed, the user should manually enter this information when the log message is requested or else make hand written notes on the hardcopy itself.
A policy that seems to work pretty well is to always start a simulation via
special init.hoc - # if the simulation uses mod file
where init.hoc xopens all necessary files to initialize the simulation.
xopen("$(NEURONHOME)/lib/hoc/noload.hoc") // standard run tools
xopen("morph.hoc") // topology, geometry, compartmentalization
xopen("memb.hoc") // membrane properties
xopen("param.hoc") // parameters that are occasionally changed
xopen("start.ses") // will automatically change for every new version
With this style, whenever an old version is checked out and run, the appearance of the screen should match the hardcopy and it should be possible to reproduce the archived simulation.
Reliance on a management policy that requires judgment is regrettable. However making the system failsafe while retaining the generality of the interpreter is probably impossible. Archive strategies consisting solely of checkpoints, eg. coredumps, use a great deal of disk space and don’t capture any semantic information about the logical structure of the variables and how they are related. This strategy is low time and high space. Audit trails, saving all the commands and changes, is high time and low space. Its problem is that it captures all the false starts and mistakes. Recapitulating every excruciating detail is probably too confusing. The current system, by developing very low space, very low time logical snapshots, although not failsafe, hopefully will realistically help to solve the very real problems of simulation management.
The following example illustrates the initialization of an essentially empty project.
Here are the files. After initializing the project with these, they can be modified to handle arbitrary complexity. These files should be in an otherwise empty subdirectory.
gnabar_hh = .120
The project is initialized with prjnrninit . This will create an RCS directory and checkout a nrnversion file with the contents:
$Revision: 1.3 $
Note that nrnversion is essentially just a manifest of the files in the project. To add a new file to the project one can explicitly check the file into RCS with the ci command and insert the appropriate line in the nrnversion file.
At this point one can run
nrniv init.hoc -
and see a neuron main menu. Use the menu to generate a graph
of an action potential. Since there was an RCS directory with a
nrnversion,v archive when nrnmainmenu() was executed,
the Miscellaneous menu has an ArchiveAndHardcopy item.
Pressing this button will archive the current version with the session
(it is saved in start.ses), request a description of this version and
print the version number, description, and session windows on the
printer specified by the
$PRINT_CMD environment variable.
It is recommended that you play with this simple project for a while to familiarize yourself with the style before employing it in a serious project. Make several different hardcopies. Use prjnrnco to check out earlier versions and run them, modify parameters, and make several more hardcopies. Note the way branch version numbers are generated and incremented.
In the current working directory creates an RCS directory and checks in (and out again with locking) all the hoc, ses, mod files as version 1.1. A nrnversion file is created (and archived) which contains a manifest of the files necessary to recreate a simulation. The Revision level of the nrnversion defines the version of the simulation.
If working files consistent with manifest in nrnversion return with exit status 0. Otherwise return with exit status 1 and print the names of the differing files on the standard output.
Checkout the version from the RCS archive.
Prior to checkout if the working files differ from the archive, the user is asked whether or not to checkout anyway. If “Checkout anyway” is chosen the changes to the previous working files will be lost.
Checkin a new version to the RCS archive.
The microemacs editor is run so the user can edit a description of the new version. On exit from the editor the user will be asked whether or not to continue checking in the version.
A new version is checked in even if the working files are unchanged.
On exit from this command, the working files are locked versions of the newest version. The nrnversion file contains the version number of the simulation (itself) as well as the version numbers of all the working files it needs to reconstruct the simulation.
cat postscriptfile | $NEURONHOME/bin/prjnrnpr
Checks in the working files and sends the postscript file to the command specified in the
$PRINT_CMD(e.g. lp) environment variable.
If the working files are not different from their archived versions the user is asked whether to continue or verify that the simulation can be reproduced. If the latter, a new the simulation is loaded in an xterm window. The user should then try to reproduce the simulation he/she is attempting to checkin. When the xterm goes away the user will be asked whether or not to continue to checkin. If you can’t reproduce the simulation or had to change working files to reproduce it, choose “Abort”
If the working files did differ then
prjnrnci()is run in an xterm.
The last question for the user to answer is whether to leave the working files at the new or old version. The answer depends on whether you envision this simulation as a side branch off the primary version or as additive.
The log message entered during checkin is added to the postcript stream and sent to
$NEURONHOME/bin/prjnrnpr "banner" "accept" "cancel"
Pops up a boolean dialog.
“1” is printed on the standard output if the “accept” button is pressed.
“0” is printed on the standard output if the “cancel” button is pressed.