GCC 15 Compiler Development Shifts From Features To Bug Fixing

[phoronix]

Phoronix: GCC 15 Compiler Development Shifts From Features To Bug Fixing

GCC 15 feature development is now officially over with entering stage three development to focus on fixing compiler bugs...

https://www.phoronix.com/news/GCC-15-Enters-Stage-3

[dmalcolm]

Note that the first part of GCC's stage 3 freeze is traditionally rather "slushy": patches that have already been posted for review may make it into GCC 15, perhaps after a little iteration.

FWIW I've got a couple myself that I posted last week that I'm hoping to get into GCC 15:

• Improvements to C++ template errors: https://gcc.gnu.org/pipermail/gcc-pa...er/668467.html; screenshot here
• Turning GCC's diagnostics subsystem into a shared library (libdiagnostics): https://gcc.gnu.org/pipermail/gcc-pa...er/668632.html

[intelfx]

Note that the first part of GCC's stage 3 freeze is traditionally rather "slushy": patches that have already been posted for review may make it into GCC 15, perhaps after a little iteration.

FWIW I've got a couple myself that I posted last week that I'm hoping to get into GCC 15:

• Improvements to C++ template errors: https://gcc.gnu.org/pipermail/gcc-pa...er/668467.html; screenshot here
• Turning GCC's diagnostics subsystem into a shared library (libdiagnostics): https://gcc.gnu.org/pipermail/gcc-pa...er/668632.html

The diagnostics thing looks gorgeous. Is it only for concepts though?

[dmalcolm]

The diagnostics thing looks gorgeous. Is it only for concepts though?

Thanks! As well as for concepts, the patch also affects the output for templates in general, and for constexpr. But you'd have to opt-in with -fdiagnostics-set-output=text:experimental-nesting=yes

[cl333r]

[LIST][*]Improvements to C++ template errors: https://gcc.gnu.org/pipermail/gcc-pa...er/668467.html; screenshot here

One of the reasons I avoided templates in C++ at all cost is that if I get an error it's almost impossible to figure out what and where's the problem cause I get 10 pages of random garbage text dumped into the terminal as the compiler's error message. Probably C++'s worst trait.

[DavidBrown]

One of the reasons I avoided templates in C++ at all cost is that if I get an error it's almost impossible to figure out what and where's the problem cause I get 10 pages of random garbage text dumped into the terminal as the compiler's error message. Probably C++'s worst trait.

This has always been a challenge for C++, and indeed any language that supports and encourages complex data or control structures. Diagnostics are to help developers find the cause of bugs in their code, and there are no rules for determining whether the bug is at the start of the first level of the hierarchy, or at the last level of the use of it. Sometimes the developer is best served by the highest level, most abstract view - such as calling a type "std::string" - and sometimes by the underlying templates, specialisations and details. Generating good diagnostics that fit what the developer needs at the time is a harder problem than parsing correct code or generating good object code in the compiler.

One of the aims of adding concepts to C++ was to simplify and clarify diagnostics, and they certainly do help. They can give a kind of a fixed point in the diagnostics - you can have errors before the concept, or after it, but good use of concepts makes it easier to find which side your problem lies on. (They have other advantages too, making code a lot clearer and easier to work with.)

And there is no doubt that gcc has been much improved here over the years. (I think competition and cooperating with clang had a good influence on them.) And some of the stuff that David Malcolm (and others) has been working on in diagnostics and static error checking is very impressive. Once his structured diagnostic data formats get supported by IDEs, it will really make things easier for developers chasing down their problems.

So if you have avoided templates in the past due to incomprehensible error messages, maybe it's time to update your tools and revisit them. If not now, then once gcc 15 is released :-)

[cl333r]

So if you have avoided templates in the past due to incomprehensible error messages, maybe it's time to update your tools and revisit them. If not now, then once gcc 15 is released :-)

Thanks, I can't run away from C++ anyway cause Qt6 uses it, but I started playing with zig and so far I like it a lot.

[dmalcolm]

So if you have avoided templates in the past due to incomprehensible error messages, maybe it's time to update your tools and revisit them. If not now, then once gcc 15 is released :-)

Thanks for the kind words.

Note that the stuff I posted above isn't yet reviewed, and even when it is, it might well be hidden behind a command-line option for GCC 15, since it's such a radical change to how we present diagnostics (I'd love to make it the default, but it needs a lot of "eating my own dogfood" before I can reasonably inflict it on end-users!)