On Fri, Sep 10, 2021 at 07:31:33AM +0900, Akira Yokosawa wrote:
> On Thu, 9 Sep 2021 10:29:52 -0700, Paul E. McKenney wrote:
> > On Thu, Sep 09, 2021 at 01:00:03PM +0900, Akira Yokosawa wrote:
> >> Hi Paul,
> >>
> >> So this patch set moves formal/data and formal/RCU-test-ratio.eps
> >> to CodeSamples/formal/data.
> >>
> >> Patch 1/2 does the movement.
> >> Patch 2/2 puts symlink to the moved .eps file under formal/.
> >>
> >> Actually, patch 2/2 is optional within the scope of this set.
> >> It presents an idea of organizing .eps files generated by gnuplot
> >> under CodeSamples/ so that redundant .eps -> .pdf conversions
> >> can be avoided.
> >>
> >> Currently, I observe three ways to include gnuplot outputs.
> >>
> >> 1) Copy generated .eps files manually to the relevant chapter's
> >> subdirectory (after running the plot script).
> >>
> >> 2) Run a plot script under CodeSamples/ and the generated .eps
> >> files are automatically copied by commands in the script.
> >>
> >> 3) Use paths of .pdf versions of .eps files under CodeSamples/...
> >> as the argument to the \includegraphics{} macros (no copying
> >> involved).
> >>
> >> Most recent plots are managed by 3).
> >
> > Good point! I moved (partially, as you noted) to #3 in part because #1
> > and #2 make it annoying to work out where the generated .eps came from.
> >
> >> However, to accommodate 3), a wildcard pattern in Makefile was
> >> extended to include deep subdirectories under CodeSamples/
> >> (commit 57b442c02db3 "Makefile: Extend wildcard pattern to catch
> >> .eps files under CodeSamples").
> >> So it now picks up a lot of unused older gnuplot plots, and causes
> >> redundant .eps -> .pdf conversions.
> >
> > And you are right, that is definitely not a good thing. And it just
> > gets worse over time as more data is regenerated.
> >
> >> So, I suggest using symlink, e.g.,
> >> formal/xxx.eps -> CodeSamples/formal/data/xxx.eps, and remove
> >> CodeSamples/ from the wildcard pattern in Makefile.
> >>
> >> That is, if it is OK to use symlink as in patch 2/2.
> >
> > I would prefer to avoid quite that many symlinks.
>
> I see.
>
> >
> >> Another advantage of this idea is that it would be easier to
> >> find out where a particular gnuplot image is originated.
> >
> > Exactly! Both #3 and symlink do that. In #3, you just look at the
> > latex for the figure, and with the symlink, just run "ls -l".
> >
> >> By copying .eps files from CodeSamples/... to other directories,
> >> you lose such origin info. This won't be a issue if there is a
> >> single .eps file of the same name under CodeSamples/, but this is
> >> not always the case.
> >>
> >> Symlinks can retain their origins.
> >>
> >> So, for example, when Zhouyi changed "rcu" to "RCU" in gnuplot
> >> generated plots (commit 2a2ca773005f "Change trace labels from
> >> "rcu" to "RCU""), he could have find out which scripts need
> >> updates more easily.
> >>
> >> As a matter of fact, he failed to update .eps files
> >> under CodeSamples/. For example:
> >>
> >> $ diff -u CodeSamples/defer/data/rcuscale.hps.2020.05.28a/refRCUperfPREEMPT.eps defer/refRCUperfPREEMPT.eps
> >> --- CodeSamples/defer/data/rcuscale.hps.2020.05.28a/refRCUperfPREEMPT.eps 2021-08-12 13:48:28.580153786 +0900
> >> +++ defer/refRCUperfPREEMPT.eps 2021-08-17 13:14:58.114517535 +0900
> >> @@ -2209,7 +2209,7 @@
> >> 1.000 UP
> >> LCb setrgbcolor
> >> 1729 1458 M
> >> -[ [(NimbusSanL-Regu) 100.0 0.0 true true 0 (rcu)]
> >> +[ [(NimbusSanL-Regu) 100.0 0.0 true true 0 (RCU)]
> >> ] -33.3 MLshow
> >> LCb setrgbcolor
> >> 1066 2212 M
> >>
> >> If Patch 2/2 looks good to you, I'll do the changes along this idea
> >> and update the wildcard pattern in Makefile.
> >
> > How about if I "git mv" the directories containing old data to some
> > name starting (say) with "OLD-", so that the hash.2013.12.02a directory
> > within CodeSamples/datastruct/hash/data becomes OLD-hash.2013.12.02a?
> > Then the Makefile could ignore .eps files in directories starting with
> > "OLD-"? And yes, would probably be an annoying change to the Makefile.
> > The only way I know to do it is to automatically generate that portion
> > of the Makefile and include it. :-/
>
> I think gnu-make's wildcard function can contain the pattern of (say):
>
> CodeSamples/*/*/OLD-*/*.eps
>
> This will match the .eps files under OLD-hash.2013.12.02a.
> A few more additional patterns can make a list of .eps files to be
> ignored.
> Then the list of latest .eps files can be made by the "filter-out"
> function.
>
> So your idea should work without that annoying change to the
> Makefile.
Even better!
> Paul, please go ahead and "git mv" to the "OLD-" directories.
I have done those in CodeSamples/datastruct/hash/data. Does that one
work the way that you had in mind?
> I will update the Makefile as a follow-up change.
Sounds good -- if the first commit looks good to you, I will do the rest.
f891fa5c6e8b ("datastruct/hash: Mark directories of unused data")
Thanx, Paul
> Thanks, Akira
>
> >
> > Or is there a better way?
> >
> > (For the moment, I will apply 1/2, and thank you!)
> >
> > Thanx, Paul
> >
> >> Thanks, Akira
> >> --
> >> Akira Yokosawa (2):
> >> formal: Move formal/data to CodeSamples/formal/data
> >> formal: Use symlink to .eps file under CodeSamples/formal/data
> >>
> >> .../formal}/data/.gitignore | 1 -
> >> CodeSamples/formal/data/RCU-test-ratio.eps | 4370 ++++++++++++++++
> >> {formal => CodeSamples/formal}/data/plot.sh | 2 -
> >> .../formal}/data/rcu-test.dat | 0
> >> formal/RCU-test-ratio.eps | 4371 +----------------
> >> 5 files changed, 4371 insertions(+), 4373 deletions(-)
> >> rename {formal => CodeSamples/formal}/data/.gitignore (50%)
> >> create mode 100644 CodeSamples/formal/data/RCU-test-ratio.eps
> >> rename {formal => CodeSamples/formal}/data/plot.sh (96%)
> >> rename {formal => CodeSamples/formal}/data/rcu-test.dat (100%)
> >> mode change 100644 => 120000 formal/RCU-test-ratio.eps
> >>
> >> --
> >> 2.17.1
> >>
