reloading this source causes Segmentation fault

[JeffreySarnoff]

I have a file that loads others, then prints examples of some type instances.
I make a few changes to any one of the files and reload the file that loads others.
After 3-5 repeats, in the middle of printing examples I get a Segmentation fault.
Restarting julia an loading the same thing runs fine, and I can run it many times.

[ViralBShah]

Could you also post an example?

[JeffreySarnoff]

Supplying the example is not easy. I load a file that loads ~ten others and then prints out about 16 lines showing typed values. Reliably, I can start julia, load the first file and when it is done displaying those values, reload it ... on the sixth time, partway through the printing to the screen there occurs a segmentation fault.

I copied selected parts of those files, those needed for essential functionality -- the selections include more than they exclude -- into a single file. I can load, run and reload that single file many times without any segmentation fault occurring. Either I happen not to have copied some problem inducing function, or the loading of a file loading more than a few others with many types and small functions gives rise to the problem. Almost everything that is called or instantiated in the multifile load is also called or instantiated in the single file version ("almost " because the only thing I know the single file does not do is a typed multiply that concatenates fields from two objects).

In my experience with the REPL, it is a fact that sometimes when I reload a single file the renewed function defs are not called, the older ones still get called (not repeatable on demand, it happens often enough but with no regularity that I have noticed). Knowing nothing of the REPL, it seems that this losing touch with some reloaded functions (keeping the old) and calling them eventually causes trouble. Maybe the fault happens the first time some part of the reload is not fully integrated into the workspace, and my repetition leads to that first occurrence.

[JeffreySarnoff]

I can put all the files in an an archive .. would that be help, where would I send it?

[JeffreySarnoff]

Warning: redefinition of constant Weeks ignored
Warning: redefinition of constant Days ignored
Warning: redefinition of constant Hours ignored
Warning: redefinition of constant Minutes ignored
Warning: redefinition of constant Seconds ignored
Warning: redefinition of constant YMDT ignored
Warning: redefinition of constant HMST ignored
Warning: redefinition of constant YMDHMST ignored
Warning: redefinition of constant chronomic ignored
Warning: redefinition of constant chrondate ignored
Warning: redefinition of constant chrontime ignored
Warning: redefinition of constant chronstep ignored
Warning: redefinition of constant calendric ignored
Warning: redefinition of constant calendate ignored
Warning: redefinition of constant calentime ignored
Warning: redefinition of constant calenstep ignored

 [ print()ed ]           [ show()n ]

      calendrical dates-of-year, times-of-day

 date-of-year:    d"2010-11-15"   "calendate(2010-11-15)"
 time-of-day:     d"14:22:55"     "calentime(14h22m55)"

      intervaled date-spans, time-spans

 date-span:         t"02-05-10"   "calendate(2010-11-15)"
 time-span:         t"16:55:10"   "calentime(14h22m55)"

        press enter << >>


 join date to time-of-day
      d"2010-11-15" * d"14:22:55" ==> d"(2010-11-15 14:22:55)"

 date+ndates,date-ndates

Segmentation fault

[pao]

You could pop them into a new git repository and push it to your Github account.

[JeffreySarnoff]

at https://github.com/jsarnoff/julia in the ./extras directory there is a file tryme.tar.gz copy that, put in the extras directory and unarchive it -- you should have a subdirectory of extras ./tm4julia. At the top level of that directory is tryme.jl, that is the file to load, after the screen printing finishes reload it, when I do that six or seven times the fault appears. The tar file has more than is run .. I did not want to miss anything.

[pao]

I meant just as plain text, not as a tarball, but that should work 😄

[JeffBezanson]

Strongly related to #265 --- code hanging around that should be removed, and some objects it depends on have been collected.

[jiahao]

is this still relevant?

[quinnj]

@JeffreySarnoff, I can't find the files to reproduce anymore, is this still an issue? Should this be subsumed by #265?

[StefanKarpinski]

I suspect that since this hasn't been reported by others since it was probably fixed – possibly inadvertently – at some point.

[quinnj]

I could imagine that if you were loading a large enough package (i.e. https://github.com/JeffreySarnoff/jtm_backup), and had a bunch of interdependencies and things were getting collected, etc. that you could get a segfault, but as you say, no one else has bumped into this, so it's probably gotten better. That's why I think if anything, we can probably just keep focused on #265 for additional improvements that may improve this kind of situation. I think we can close this now though.

[StefanKarpinski]

Agreed.