-project, replicate- and .ster files.

[Jorge Eduardo Perez Perez]

Dear Statalist.

I use Robert Picard 's incredibly useful -project- command frequently. My workflow usually goes from datasets, to .ster files with estimation results, to tables and figures.

I use -project, replicate- to check that my projects can be replicated from start to end. This basically runs the projects twice and checks the files for differences. I have noticed that the .ster files are always different between replications, even if the results have not changed. The files are checked for differences with the -checksum- command.

Why would this be the case? Do .ster files include time-specific information? Robert Picard Is there a way to make -project- to compare .ster files the same way it compares log files, ignoring time-specific information?

Thanks,

[Robert Picard]

Stata's .ster files indeed contain time stamps:

. sysuse auto, clear
(1978 Automobile Data)

. quietly regress mpg weight displ if foreign

. estimates save myreg.ster
file myreg.ster saved

. type myreg.ster, starbang
*! Stata(R) 015.10 Binary File Header_version 0010
*! Date 07may2019 09:16:17                                      
*! Byteorder LOHI
*! Filetype estimates                        Version 0006
*! <end_of_header>
*! Date 07may2019 09:16:17                                      
*! Byteorder LOHI
*! Filetype estimates                        Version 0001
*! <end_of_header>

.

Obviously, these files will differ at every replication build. I'm not particularly interested in making additional exceptions in project for files that Stata chooses to bless with a file stamp. I guess that one option is to just ignore lines in the replication report that indicate a change if they relate to .ster files. Another is to not declare dependencies for these files. These files are not results so for the purpose of a replication build, it should not matter.

A more sophisticated approach is to set a global flag that you would use to indicate that you are performing a replication build. At the top level of the project's directory, you create a text file that contains:

local replicate 1

if you are about to perform a replication build or

local replicate 0

if not. Say this text file is called "replicate.txt". You can then, in any do-file, have commands run conditionally:

* top of do-file
project, doinfo
local pdir "`r(pdir)'"                            // the project's main dir.
local dofile "`r(dofile)'"                        // do-file's stub name
project, original("`pdir'/replicate.txt")
include "`pdir'/replicate.txt"

* eventually save one or more estimates
estimates save "myreg1.ster"
if !`replicate' project, creates("myreg1.ster")

I use this technique when I use timers in my do-files. Since the timings change slightly at each run, I list them conditionally:

if !`replicate' timer list

[Jorge Eduardo Perez Perez]

This is very useful, thanks. I'll probably just ignore the lines associated with .ster files in the replication report for my existing projects.