If there is a good, general, alternative approach for this, we can support it in Bisect_ppx. Unfortunately, I don’t know enough about Menhir to be able to propose anything specific at this point.

Best,

Anton

On Mar 6, 2016, at 16:53, Gabriel Scherer <gabriel.scherer@gmail.com> wrote:

This is an interesting question and, as far as I know, there is no good solution using existing versions of the interacting tools.

Below very simple patch that will add (*BISECT-IGNORE*) in front of every line of code generated by Menhir, except those written by the programmer (the "strecthes" in Menhir-speak). It applies cleanly on top of the latest released Menhir archive,
  http://gallium.inria.fr/~fpottier/menhir/menhir-20160303.tar.gz

The patch as-is is obviously a hack: it would need to be a configuration option when running menhir, and hard-coding Bisect (or bisect_ppx)'s syntax into Menhir is not elegant. One could try to have a configuration option to let users write a fixed string (or comment) at the beginning of each generated code line, but I'm not sure whether François Pottier (in cc:) would consider this is elegant enough. François, would you comment on whether this is a direction that seems acceptable to you?

(Bisect support ignoring entire regions at once by using (*BISECT-IGNORE-BEGIN*) and (*BISECT-IGNORE-END*); we could try to implement that instead of a per-line change, but I suspect that it would be slightly harder to implement (you have to hook the beginning of input, end of input, and around each user-code insertion) for no real gain.)

Toggling code-coverage semantics by inserting comments is not a very nice interface (although rather logical when you think of the level of generality required), so it's a bit frustrating that parser generators would have to play at this level. It would be better to have a more structured, unified interface supported by all the code-coverage tools, but to my knowledge no such thing exists.


From d595ba5149a314c56623e1735af7678f5f62d525 Mon Sep 17 00:00:00 2001
From: Gabriel Scherer <gabriel.scherer@gmail.com>
Date: Sun, 6 Mar 2016 17:43:14 -0500
Subject: [PATCH] output (*BISECT-IGNORE*) in front of each
 non-programmer-written line

EXPERIMENTAL PATCH: this should of course be turned into an explicit option
---
 src/printer.ml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/src/printer.ml b/src/printer.ml
index ea978bc..714bb08 100644
--- a/src/printer.ml
+++ b/src/printer.ml
@@ -46,6 +46,7 @@ let rawnl f =
 
 let nl f =
   rawnl f;
+  output_string f "(*BISECT-IGNORE*)";
   output_substring f whitespace 0 !indentation
 
 let indent ofs producer f x =
--
2.5.0

On Sun, Mar 6, 2016 at 2:53 PM, Vincent Jacques <vincent@vincent-jacques.net> wrote:

Hello,

Does somebody have experience measuring test coverage of generated lexers/parsers?

I'm using ocamllex/ocamlyacc [1] (but I can switch to Menhir [2]) to generate a lexer/parser. In my tests, I simply check that some input strings give the ASTs I expect.

I usually use Bisect [3] to make sure that my tests cover the code I intended to cover, but in that configuration, Bisect is lost between the .mll/.mly files and the generated .ml files and produces useless reports.

How would you measure test coverage in that case?

Thanks,

[1] http://caml.inria.fr/pub/docs/manual-ocaml/lexyacc.html
[2] http://gallium.inria.fr/~fpottier/menhir/
[3] http://bisect.x9c.fr/
--

Vincent Jacques
http://vincent-jacques.net

"S'il n'y a pas de solution, c'est qu'il n'y a pas de problème"
            Devise Shadock

<0001-output-BISECT-IGNORE-in-front-of-each-non-programmer.patch>