I'm trying to compile druntime for a custom OS. When I try to compile for the target triple -target=arm-none-eabi, the compiler does nothing without any error messages and there is no object file output at all. However according to this guide.

https://wiki.dlang.org/Minimal_semihosted_ARM_Cortex-M_%22Hello_World%22

it uses the triple thumb-none-linux-eabi, which is kind odd of but it might work for that particular case.

The problem is when using thumb-none-linux-eabi, the compiler automatically inserts D version identifiers like Posix. There are ways around this like putting your desired target sooner in the version if-else statements but still for completeness shouldn't the non OS target be supported and why aren't there any warning messages when using the arm-none-eabi target?

On Friday, 13 March 2020 at 14:41:03 UTC, IGotD- wrote:
> The problem is when using thumb-none-linux-eabi, the compiler automatically inserts D version identifiers like Posix.

Yes, Linux is a Posix target.

> but still for completeness shouldn't the non OS target be supported

It is and doesn't predefine Posix etc. then; try `-mtriple=thumb--none-eabi` (or `thumb-unknown-eabi` etc.).

> and why aren't there any warning messages when using the arm-none-eabi target?

Probably because it works just fine, at least it does for me with LDC v1.20 on Windows - `ldc2 -c -mtriple=arm-none-eabi hello.d -v && ls -l hello.o`. `-v` shows the expanded triple (3rd line, => arm-none-unknown-eabi).

On Friday, 13 March 2020 at 17:06:03 UTC, kinke wrote:
> On Friday, 13 March 2020 at 14:41:03 UTC, IGotD- wrote:
>> The problem is when using thumb-none-linux-eabi, the compiler automatically inserts D version identifiers like Posix.
>
> Yes, Linux is a Posix target.
>
>> but still for completeness shouldn't the non OS target be supported
>
> It is and doesn't predefine Posix etc. then; try `-mtriple=thumb--none-eabi` (or `thumb-unknown-eabi` etc.).
>
>> and why aren't there any warning messages when using the arm-none-eabi target?
>
> Probably because it works just fine, at least it does for me with LDC v1.20 on Windows - `ldc2 -c -mtriple=arm-none-eabi hello.d -v && ls -l hello.o`. `-v` shows the expanded triple (3rd line, => arm-none-unknown-eabi).

You are right, any other file works with arm-none-eabi.

Actually this might be more related to druntime rather than arm-none-eabi.

Try.

cd ldc/runtime/druntime/src (where your druntime is located).
ldc2 -c --output-o -conf= -w -de -dip1000 -mtriple=arm-none-eabi -O3 -release -preview=fieldwise -od=/tmp/objects -op core/atomic.d

ls /tmp/objects/core, and there is no atomic.o file.

the compiler just gives a warning: Warning: Assuming critical section size = 40 bytes
not sure what this is about.

change -mtriple to arm-none-linux-eabi and the object file is created.

I've tried to find if some identifier is missing because of this but I haven't found any also atomic should only be CPU architecture dependent.

On Friday, 13 March 2020 at 21:47:02 UTC, IGotD- wrote:
> cd ldc/runtime/druntime/src (where your druntime is located).
> ldc2 -c --output-o -conf= -w -de -dip1000 -mtriple=arm-none-eabi -O3 -release -preview=fieldwise -od=/tmp/objects -op core/atomic.d

Works fine for me, again with v1.20 on Windows, no warnings wrt. unknown section size (needs OS-specific compiler support for `synchronized`). I can only guess that you are using an older LDC version.

On Friday, 13 March 2020 at 22:11:57 UTC, kinke wrote:
> I can only guess that you are using an older LDC version.

Note that druntime and the compiler are tightly coupled because both depend on each other, so you should always use the corresponding LDC version to compile druntime. I.e., not trying to build our current druntime (branch `ldc`) with LDC v1.20, nor trying to build druntime v1.20/2.090 with older LDC versions. ldc-build-runtime usually handles that for you. - The atomic intrinsics have been reworked recently.

On Saturday, 14 March 2020 at 12:51:30 UTC, kinke wrote:
>
> Note that druntime and the compiler are tightly coupled because both depend on each other, so you should always use the corresponding LDC version to compile druntime. I.e., not trying to build our current druntime (branch `ldc`) with LDC v1.20, nor trying to build druntime v1.20/2.090 with older LDC versions. ldc-build-runtime usually handles that for you. - The atomic intrinsics have been reworked recently.

I'm using LDC 1.20.1. I cloned the entire LDC git for my changes in the runtime and the version for druntime in particular is e96b6f2.

Date:   Tue Feb 25 01:50:21 2020 +0100

    ldc.intrinsics: Add support for LLVM 11 (#173)

Latest tag is ldc-v1.20.0 three commits earlier and I don't find any tag for 1.20.1 so my git might be a bit old.

On Friday, 13 March 2020 at 17:06:03 UTC, kinke wrote:
>
> Probably because it works just fine, at least it does for me with LDC v1.20 on Windows

I just tried with LDC 1.20.1 on Windows, there it compiles the to the object files just as you previously tested. We have a difference between Linux and Windows for the same command line.

On Wednesday, 18 March 2020 at 12:57:26 UTC, IGotD- wrote:
> On Friday, 13 March 2020 at 17:06:03 UTC, kinke wrote:
>>
>> Probably because it works just fine, at least it does for me with LDC v1.20 on Windows
>
> I just tried with LDC 1.20.1 on Windows, there it compiles the to the object files just as you previously tested. We have a difference between Linux and Windows for the same command line.

The difference is that on Windows hosts, we don't have any fallback value for the critical section size for an unknown; the first time it is needed, an error is issued.

For Posix hosts, we use the native `pthread_mutex_t` size as fallback value, coupled with a one-time warning issued during initialization. The `-w` in your cmdline aborts on warnings; you can use `-wi` to have them printed but continue.

It might be better never to fall back to the native size, not just for consistency across hosts.

On Friday, 13 March 2020 at 14:41:03 UTC, IGotD- wrote:
> I'm trying to compile druntime for a custom OS. When I try to compile for the target triple -target=arm-none-eabi, the compiler does nothing without any error messages and there is no object file output at all. However according to this guide.
>
> https://wiki.dlang.org/Minimal_semihosted_ARM_Cortex-M_%22Hello_World%22
>
> it uses the triple thumb-none-linux-eabi, which is kind odd of but it might work for that particular case.
>
> The problem is when using thumb-none-linux-eabi, the compiler automatically inserts D version identifiers like Posix. There are ways around this like putting your desired target sooner in the version if-else statements but still for completeness shouldn't the non OS target be supported and why aren't there any warning messages when using the arm-none-eabi target?

IMO, ldc right now will not work without a real OS like windows/linux/macos, as druntime and phobos is built upon with clib. And none-eabi doesn't have the "standard C library".

it will only work with arm-none-eabi target when trying to build LDC a "betterC compiler".

On Tuesday, 31 March 2020 at 09:59:44 UTC, dangbinghoo wrote:
> On Friday, 13 March 2020 at 14:41:03 UTC, IGotD- wrote:
>
> IMO, ldc right now will not work without a real OS like windows/linux/macos, as druntime and phobos is built upon with clib. And none-eabi doesn't have the "standard C library".
>
> it will only work with arm-none-eabi target when trying to build LDC a "betterC compiler".

I'm using arm-none-eabi so that the compiler doesn't pre-select any operating system. Then I just use the --d-version option to inject my own verison of my custom OS. This seems to be working