37 comments

[ 3.1 ms ] story [ 49.9 ms ] thread
It's worth celebrating (well, noting in an approving manner) this release. GCC is a foundation stone project for, well, a huge amount of the world's wealth.

It is a healthy project (even with CLang etc), and with the heartbleed post mortem on our minds we should probably ask what can be done to keep it and others healthy - auditing, foundation status etc.

So thank you GCC developers, for making sure I do not need to worry about something.

Bit strange comment; there are still plenty of bugs in GCC to worry about.
Any benchmarks? How does this compare with previous versions of gcc?
The release notes mention several benchmark comparisons.
I don't see any benchmarks showing improvements in 5.2. The 5.2 information seems sparse. Could you point this out?
You should investigate by yourself :)
Does GCC support modules? http://clang.llvm.org/docs/Modules.html
gfortran which is part of GCC supports modules (this is because Fortran the language supports modules).

If you meant modules for C or C++, no (as far as I know), probably because C and C++ have no standard for modules. In a recent talk Bjarne Stroustroup talks about his hope that C++17 will include support for modules https://www.youtube.com/watch?v=2egL4y_VpYg

It probably will once the C++ committee figures out how they want to do it...
> Write-only variables are now detected and optimized out.

That's probably something one should keep in mind when writing micro-benchmarks.

There should be a warning for this enabled in every C project.
GCC has a warning for that, and yes, it should be.
(comment deleted)
Those pointers always need to be defined as volatile anyway
(comment deleted)
The point of using volatile is to make sure not to put the variable in a register or use a cached value; it hits the memory.
(comment deleted)
I'm appalled that this wasn't taken care of 20 years ago. It seems like such low-hanging fruit.
This is a standard compiler feature, yes -- lots of real code out there has a lot of dead writes, such as debugging code that has been only partly removed or disabled. Historically, it's been a significant win for some of the SPECcpu benchmarks.
Historically, it's been a significant win for some of the SPECcpu benchmarks.

But doesn't it mean that you're not really executing the full benchmark then?

No, because no one ever reads the dead data. These are not microbenchmarks that get broken by optimizations like this; SPECcpu programs have answers, and the goal is to output the answer. The answer is tested. Anything you can delete from the program is fair game.
It's standard to remove local variables that are write-only, but globals aren't always defined to be unused even when you want them to be. If you're building a shared library and you exported the symbol, it has to leave it in as part of the API.

Link-time optimization can delete a lot more stuff when you're building a program directly, since it only has to keep main()… as long as you didn't go and use things like function pointers.

This is one reason I try to use '-fvisibility=hidden' in libraries, because it prevents anything from being part of the API unless you go back and specifically export it.

I assume it means globals, which is significantly less trivial. Any SSA-based compiler, which GCC has been for years and years, trivially does this for locals.
It's worth noting that, due to GCC's new and esoteric version numbering scheme[1], this is a bugfix-only release on the 5.x branch. The bigger news was when GCC 5.1 (a major stable release) was released in April. That said, GCC 5.2 fixes a pretty serious bug that I was hitting so I'm happy to hear about it!

[1] https://gcc.gnu.org/develop.html#num_scheme

They switched the default C compilation mode from --std=gnu89 to --std=gnu11 in a bugfix-only release?
No, they did that in 5.1. The linked page is mostly about the overall 5.x series, with a tiny little section at the bottom about 5.2.
I hope more projects start using it (or C99) over C89.
Woah, new and esoteric indeed!

    X.0.0 is experimental development
    X.0.1 is pre-release stabilization stage
    X.1.0 is the stable release
    X.1.1 is development of release branch (fixes and backports)
    X.2.0 is the second stable release
It looks like only X.1.0, X.2.0, etc uniquely tag a single point in the development history. Are all the others just moving labels?
With this release the golang frontend catches up it'so Go version 1.4.2.

EDIT: I'm wrong. That was GCC 5.1.

Nopes. Go 1.4.2 was released with GCC 5.1 (April 22, 2015). This release is just bugfixes.
Thanks; you're right. I read the 5 series general release notes as if it were for 5.2.
I'm more interested in why GCC still doesn't have a D frontend.

Back when people were criticizing Google for acquiring GCC developers and then adding their own Go language I think they promised to add D in like 4.6 or something like that.

Hi guys, any idea when we might get this on mingw?
Where is GCC's place in the world now that there is LLVM/Clang?

Does the LLVM+Clang combo make GCC obsolete?