http://gcc.gnu.org/bugzilla/show_bug.cgi?id=44054
--- Comment #5 from Manuel López-Ibáñez <manu at gcc dot gnu.org> --- Created attachment 32622 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=32622&action=edit proof of concept The attached patch is a proof of concept. It gives the following output: /home/manuel/test1/test-gfc-warning.f03:11:0: Warning: Possible change of value in conversion from INTEGER(8) to INTEGER(4) at (1) [-Wconversion] arr = (/ INTEGER(KIND=4) :: HUGE(0_8) /) ! { dg-warning "conversion from" } ^ This warning is controllable by -Wno-conversion, -Werror=conversion and the #pragma GCC diagnostic. Also the diagnostic is in color (unfortunately it does not show it here). Contrary to what I said above, I think it is actually easier for Fortran to use the common pretty printer rather than fake their own. It will also remove quite a bit of duplicated code from the Fortran FE. The transition can be done in the following steps (I chose the wrong warning to start with): 1. Warnings that don't use %L (so initially you don't have to deal with locus) 2. Warnings that use %C 3. Warnings that use one %L 4. Warnings that use two %L If there is one or more willing Fortran developers that wants to tackle this for the next release, I think the four steps are feasible within one release. To fully match the current Fortran diagnostics, the common diagnostics machinery needs the following features: a. Make the printing of the caret line more customizable. For the C/C++ FE it will be moved to the diagnostic_finalizer, whereas for Fortran it will go in the diagnostic_starter, so it can be printed before the actual text. b. A way to customize the caret symbol so Fortran can print '1' instead of '^'. c. Pass down offsets relative to current location. This is anyway necessary for printing more precise diagnostics for format strings, so it is a matter to add an offset field to the diagnostic_info structure. d. A way to print two locations in the caret line. And that is all as far as I can see. If a Fortran developer volunteers to tackle the Fortran bits, I will be happy to implement the common diagnostics bits. From the above a) and b) are trivial. c) is also trivial but requires writing a bit more of code. d) is a bit more convoluted but it should work fine. The Fortran transition can be incremental, that is, if Fortran disables colors and -fdiagnostics-show-option by default, then the converted diagnostics will be indistinguishable from the unconverted ones (except that -Wno-*, -Werror= and the #pragmas will only work for the converted ones).
