two (short) new entries for Rcpp-FAQ vignette

[eddelbuettel]

Two simple additions which provide a start.

@kevinushey @jjallaire if you have comments / extensions maybe commit straight into branch?

[jjallaire]

I think this might in it's present form be too cautionary. In all Rcpp attribute generated code (including the export of C++ interfaces to other packages) we handle this gracefully. The only place I'm aware of in Rcpp where this bites is specifically in Rcpp Module methods on Windows only. I think we might be better off just adding docs to modules saying you can't call stop() from within methods (they should always return something even if just an error object).

[eddelbuettel]

Hm, I thought we still bound by the general restriction of not being able to throw/catch across distinct DLLs? But maybe I am indeed to pessimistic in that view as it would blow up the use case of, eg, RQuantLib catching QuantLib errors.

Easy to rewrite as "Mostly. A known remaining limitation ..." I can take a stab at it later.

[kevinushey]

Cross-DLL exceptions work as long as:

1. You are linking to a non-static C++ standard library;
2. You are using the same standard library in all compiled objects.

Unfortunately, 1. is not true on Windows as R opts to link to libstdc++ statically; there was some discussion about whether this could be changed for Rtools 4.0 but it sounds like that won't happen.

In some cases, 2. is also not true -- e.g. on macOS users might end up mixing the version of libc++ bundled with R in some dylibs, and the system version of libc++ in others. Usually this means that the exception type info isn't properly handled across module boundaries.

[kevinushey]

(and of course, Rcpp attributes help us as, if one calls an attributes-generated function, any C++ exceptions are effectively shielded by try-catch and so are not propagated into a separate DLL)

[eddelbuettel]

@jjallaire @kevinushey Thanks for the very helpful comments.

I just rebased and force-pushed this branch adding one refining commit. The diff in the commit now shows an expanded and refined answer to this new FAQ entry based on your comments.

[jjallaire]

I'm not sure about the RQuantLib case but I'd say given we haven't really heard about this save for modules that perhaps there is some other mediation in place there?

…

On Mon, Oct 28, 2019 at 7:38 AM Dirk Eddelbuettel ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In vignettes/rmd/Rcpp-FAQ.Rmd <#1009 (comment)>: > @@ -1720,3 +1737,11 @@ elsewhere. So if your autogenerated file fails, and a `symbols not found` error is reported by the linker, consider running `compileAttributes()` twice. Deleting `R/RcppExports.R` and `src/RcppExports.cpp` may also work. + +## Can we use exceptions and stop() across shared libraries? + +No, sadly. A known limitation coming from the operating system, Windows in Hm, I thought we still bound by the general restriction of not being able to throw/catch across distinct DLLs? But maybe I am indeed to pessimistic in that view as it would blow up the use case of, eg, RQuantLib catching QuantLib errors. Easy to rewrite as "Mostly. A known remaining limitation ..." I can take a stab at it later. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1009?email_source=notifications&email_token=AAAZPRYG7ST25WCAF43AQ43QQ3FNDA5CNFSM4JFY4WA2YY3PNVWWK3TUL52HS4DFWFIHK3DMKJSXC5LFON2FEZLWNFSXPKTDN5WW2ZLOORPWSZGOCJMHYYI#discussion_r339518406>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAZPR5PQMRTEEXLWRS24QLQQ3FNDANCNFSM4JFY4WAQ> .

[eddelbuettel]

Yes I meant that in general it just works (at least on not-on-Windows). I.e. at work we have a large-ish standard C++ codebase, exceptions are generally forbidden and it asserts and exits on error so for the calling-from-R-case I redefine these as Rcpp::stop() and it works just fine.

Just not sure about Windows which I have the benefit of mostly not seeing anymore ....

[codecov-io]

Codecov Report

Merging #1009 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #1009   +/-   ##
=======================================
  Coverage   82.46%   82.46%           
=======================================
  Files          63       63           
  Lines        3166     3166           
=======================================
  Hits         2611     2611           
  Misses        555      555

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ee0a5b8...05a1d51. Read the comment docs.

[eddelbuettel]

@kevinushey @jjallaire if/when you have a chance glance over the current state of the new paragraphs and especially the updated / rewritten second one reflecting your really useful feedback.

Otherwise looks like we are good to go. May toss this at CRAN soon as 1.0.3.

[eddelbuettel]

Ok, I will take "no further comment" in the generous sense of "no further complaints" 😀

Merging now and going to prepare a 1.0.3 release.