Menu
Search

Forums

Forum Index

Thread overview
Disabling All Inlining in DMD Debug Builds
Oct 24, 2022
Jack Stouffer
Oct 24, 2022
Imperatorn
Oct 24, 2022
ryuukk_
Oct 25, 2022
Per Nordlöw
Oct 25, 2022
ryuukk_
Oct 25, 2022
Jack Stouffer
October 24, 2022
Gravatar of Jack StoufferPosted by Jack StoufferReply
Jack Stouffer
Gravatar of Jack Stouffer


Reply

I use

pragma(inline, true)
function definition

all over my code. And by default, DMD inlines these functions even in debug builds, which normally is great. I have a custom dynamic array container and if the indexing operator overload function wasn't inlined the debug build would have a ~10x slowdown and become un-testable in many ways.

What sucks about this is that stepping through the code in a debugger is way worse because the the "step-over" operation no longer works properly. Maybe this is a bug in the PDB output, but trying to step over an inlined function call still takes you into the inlined function body as if it were a step-in.

So is there a way to tell DMD to ignore these pragmas in certain builds?

October 24, 2022
Gravatar of ImperatornPosted by Imperatorn
in reply to Jack Stouffer		Reply
Imperatorn
Gravatar of Imperatorn
Posted in reply to Jack Stouffer


Reply

On Monday, 24 October 2022 at 19:28:34 UTC, Jack Stouffer wrote:

>

I use

pragma(inline, true)
function definition

all over my code. And by default, DMD inlines these functions even in debug builds, which normally is great. I have a custom dynamic array container and if the indexing operator overload function wasn't inlined the debug build would have a ~10x slowdown and become un-testable in many ways.

What sucks about this is that stepping through the code in a debugger is way worse because the the "step-over" operation no longer works properly. Maybe this is a bug in the PDB output, but trying to step over an inlined function call still takes you into the inlined function body as if it were a step-in.

So is there a way to tell DMD to ignore these pragmas in certain builds?

Search and replace pragma(inline, true) with //pragma(inline, true) ;)

October 24, 2022
Gravatar of ryuukk_Posted by ryuukk_
in reply to Jack Stouffer		Reply
ryuukk_
Gravatar of ryuukk_
Posted in reply to Jack Stouffer


Reply

On Monday, 24 October 2022 at 19:28:34 UTC, Jack Stouffer wrote:

>

I use

pragma(inline, true)
function definition

all over my code. And by default, DMD inlines these functions even in debug builds, which normally is great. I have a custom dynamic array container and if the indexing operator overload function wasn't inlined the debug build would have a ~10x slowdown and become un-testable in many ways.

What sucks about this is that stepping through the code in a debugger is way worse because the the "step-over" operation no longer works properly. Maybe this is a bug in the PDB output, but trying to step over an inlined function call still takes you into the inlined function body as if it were a step-in.

So is there a way to tell DMD to ignore these pragmas in certain builds?

This is ugly, but that works

version(DebugFast) pragma(inline, true);
void my_function()
{
}

But that'll only inline if you build with DEBUG_FAST, wich is not what you want

I wish we could do version(DebugFast | Release)

October 25, 2022
Gravatar of Per NordlöwPosted by Per Nordlöw
in reply to ryuukk_		Reply
Per Nordlöw
Gravatar of Per Nordlöw
Posted in reply to ryuukk_


Reply

On Monday, 24 October 2022 at 20:43:45 UTC, ryuukk_ wrote:

>

I wish we could do version(DebugFast | Release)

Moreover, I've been using, for instance,

version(D_Coverage) {} else pragma(inline, true);
void foo() {}

to get correct coverage inclusion of foo(). Is this still neeed?

October 25, 2022
Gravatar of ryuukk_Posted by ryuukk_
in reply to Per Nordlöw		Reply
ryuukk_
Gravatar of ryuukk_
Posted in reply to Per Nordlöw


Reply

On Tuesday, 25 October 2022 at 08:14:26 UTC, Per Nordlöw wrote:

>

On Monday, 24 October 2022 at 20:43:45 UTC, ryuukk_ wrote:

>

I wish we could do version(DebugFast | Release)

Moreover, I've been using, for instance,

version(D_Coverage) {} else pragma(inline, true);
void foo() {}

to get correct coverage inclusion of foo(). Is this still neeed?

The 'else' is easy to miss when reading that code

There must be a better solution than that, version is just too rigid

Looks like OP opened an issue about it: https://issues.dlang.org/show_bug.cgi?id=23435

October 25, 2022
Gravatar of Jack StoufferPosted by Jack Stouffer
in reply to Per Nordlöw		Reply
Jack Stouffer
Gravatar of Jack Stouffer
Posted in reply to Per Nordlöw


Reply

On Tuesday, 25 October 2022 at 08:14:26 UTC, Per Nordlöw wrote:

>

On Monday, 24 October 2022 at 20:43:45 UTC, ryuukk_ wrote:

>

I wish we could do version(DebugFast | Release)

Moreover, I've been using, for instance,

version(D_Coverage) {} else pragma(inline, true);
void foo() {}

to get correct coverage inclusion of foo(). Is this still neeed?

This looks like a workable solution for now. Thanks