www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - removal of dead functions and methods

reply ketmar via Digitalmars-d <digitalmars-d puremagic.com> writes:
Hello.

i was never thinking about it, but recently i found parts of source
code in my compiled D binary. they comes from code generation
functions, which only used in compile time, yet still happily sits in
resulting binary. are there any plans to somehow "fix" that?

i know that this is more a linker issue, but having unused code with
unused parts of the source in resulting binary can be undesirable
sometimes.

sure, we can "fix" this by using "functional templates" instead of CTFE
functions, but this is not always handy. and i think that "functional
template programming" consumes more memory than CTFE (yet i didn't
check this).

for now just remember that strings from your CTFE-only functions can
get into resulting binary, so don't write fun comments about your
customers there. ;-)
Dec 24 2014
parent reply "Dicebot" <public dicebot.lv> writes:
On Wednesday, 24 December 2014 at 14:04:56 UTC, ketmar via 
Digitalmars-d wrote:

 Hello.

 i was never thinking about it, but recently i found parts of 
 source
 code in my compiled D binary. they comes from code generation
 functions, which only used in compile time, yet still happily 
 sits in
 resulting binary. are there any plans to somehow "fix" that?


LDC fixes that by using garbage collection of unused symbols by 
default.
Dec 24 2014
next sibling parent reply ketmar via Digitalmars-d <digitalmars-d puremagic.com> writes:
On Wed, 24 Dec 2014 14:08:23 +0000
Dicebot via Digitalmars-d <digitalmars-d puremagic.com> wrote:


 On Wednesday, 24 December 2014 at 14:04:56 UTC, ketmar via=20
 Digitalmars-d wrote:

 Hello.

 i was never thinking about it, but recently i found parts of=20
 source
 code in my compiled D binary. they comes from code generation
 functions, which only used in compile time, yet still happily=20
 sits in
 resulting binary. are there any plans to somehow "fix" that?


=20
 LDC fixes that by using garbage collection of unused symbols by=20
 default.


yet we have two other compilers too. i didn't checked what GDC does
with "--gc-sections" cli switch, but i think that it doesn't matter.
and DMD pollutes binaries.

and i think that DMD is most widely-used D compiler nowdays.

btw: i can't check it, but does LDC removing unused static methods too?
i think it should do this, as they aren't very different from free
functions internally.
Dec 24 2014
parent reply "Dicebot" <public dicebot.lv> writes:
On Wednesday, 24 December 2014 at 14:15:40 UTC, ketmar via 
Digitalmars-d wrote:

 yet we have two other compilers too. i didn't checked what GDC 
 does
 with "--gc-sections" cli switch, but i think that it doesn't 
 matter.


GDC is not there yet but I remember someone working on it. It 
needs tweaks to runtime symbol emitting.


 and DMD pollutes binaries.


DMD is not much suitable for anything but experiments and 
debugging anyway.


 and i think that DMD is most widely-used D compiler nowdays


Those who care about generated binaries don't use DMD, that simple
Dec 24 2014
next sibling parent ketmar via Digitalmars-d <digitalmars-d puremagic.com> writes:
On Wed, 24 Dec 2014 14:20:24 +0000
Dicebot via Digitalmars-d <digitalmars-d puremagic.com> wrote:


 DMD is not much suitable for anything but experiments and=20
 debugging anyway.


i was thinking the same, but i found that code produced by DMD is
acceptable in most cases. there is no much need in thightly optimized
code for most tasks, as tasks tend to be either bound by I/O or by
user. i found that even some real-time graphics is ok with DMD.

so i can't agree with you on this matter.


 and i think that DMD is most widely-used D compiler nowdays


 Those who care about generated binaries don't use DMD, that simple


besides, it's not always possible to use GDC or LDC, as they tend to be
off by at least one frontend/phobos version, or have some unfixed bugs
that prevents building binary.
Dec 24 2014
prev sibling next sibling parent Joseph Rushton Wakeling via Digitalmars-d <digitalmars-d puremagic.com> writes:
On 24/12/14 15:35, ketmar via Digitalmars-d wrote:

 besides, it's not always possible to use GDC or LDC, as they tend to be
 off by at least one frontend/phobos version, or have some unfixed bugs
 that prevents building binary.


That's something that is better fixed by adjusting the design of the frontend 
such that a new version can be dropped plug-and-play style on top of an 
arbitrary backend.
Dec 25 2014
prev sibling next sibling parent Iain Buclaw via Digitalmars-d <digitalmars-d puremagic.com> writes:
On 25 December 2014 at 16:49, Joseph Rushton Wakeling via
Digitalmars-d <digitalmars-d puremagic.com> wrote:

 On 24/12/14 15:35, ketmar via Digitalmars-d wrote:

 besides, it's not always possible to use GDC or LDC, as they tend to be
 off by at least one frontend/phobos version, or have some unfixed bugs
 that prevents building binary.



 That's something that is better fixed by adjusting the design of the
 frontend such that a new version can be dropped plug-and-play style on top
 of an arbitrary backend.



First, there needs to be better ownership of tasks done in the
frontend vs. backend.  Who error's about what? Who rewrites codegen?
There's currently far too many moving of functionality/heuristics
between both to allow this sort of plug-and-play style.  For instance,
see [1].

But even if it were more pluggable, you can't expect the interface to
remain constant - but at least more manageable. :)

[1]: https://github.com/D-Programming-Language/dmd/pull/4062
Dec 25 2014
prev sibling parent ketmar via Digitalmars-d <digitalmars-d puremagic.com> writes:
On Thu, 25 Dec 2014 17:49:36 +0100
Joseph Rushton Wakeling via Digitalmars-d <digitalmars-d puremagic.com>
wrote:


 On 24/12/14 15:35, ketmar via Digitalmars-d wrote:

 besides, it's not always possible to use GDC or LDC, as they tend to be
 off by at least one frontend/phobos version, or have some unfixed bugs
 that prevents building binary.


=20
 That's something that is better fixed by adjusting the design of the fron=


tend=20

 such that a new version can be dropped plug-and-play style on top of an=20
 arbitrary backend.


i'm in no way telling that LDC and GDC authors doing something wrong.
i'm just admitting the fact. i'll be very happy if GDC will be on par
with DMD, and if D frontend will become "stable" in the means of API,
so switching to new version will not take month of manwork.

it seems to me that with introducing visitors everywhere future frontend
code will be easier to integrate.
Dec 25 2014
prev sibling parent reply "Kapps" <opantm2+spam gmail.com> writes:
On Wednesday, 24 December 2014 at 14:08:25 UTC, Dicebot wrote:

 On Wednesday, 24 December 2014 at 14:04:56 UTC, ketmar via 
 Digitalmars-d wrote:

 Hello.

 i was never thinking about it, but recently i found parts of 
 source
 code in my compiled D binary. they comes from code generation
 functions, which only used in compile time, yet still happily 
 sits in
 resulting binary. are there any plans to somehow "fix" that?


 LDC fixes that by using garbage collection of unused symbols by 
 default.


Does DMD/Optlink on Windows do this too? I've noticed that using 
DMD/Optlink on Windows, my executable is 4MB for unittests on a 
module that uses a lot of CTFE template methods, while with 
DMD/ld on Linux (and I believe GDC/ld as well), the same test 
executable is over 40MB.
Dec 24 2014
next sibling parent Walter Bright <newshound2 digitalmars.com> writes:
On 12/24/2014 7:10 AM, Kapps wrote:

 Does DMD/Optlink on Windows do this too? I've noticed that using DMD/Optlink on
 Windows, my executable is 4MB for unittests on a module that uses a lot of CTFE
 template methods, while with DMD/ld on Linux (and I believe GDC/ld as well),
the
 same test executable is over 40MB.


You can find out what is in an executable by linking with the /MAP option.
Dec 24 2014
prev sibling parent ketmar via Digitalmars-d <digitalmars-d puremagic.com> writes:
On Wed, 24 Dec 2014 15:10:17 +0000
Kapps via Digitalmars-d <digitalmars-d puremagic.com> wrote:


 On Wednesday, 24 December 2014 at 14:08:25 UTC, Dicebot wrote:

 On Wednesday, 24 December 2014 at 14:04:56 UTC, ketmar via=20
 Digitalmars-d wrote:

 Hello.

 i was never thinking about it, but recently i found parts of=20
 source
 code in my compiled D binary. they comes from code generation
 functions, which only used in compile time, yet still happily=20
 sits in
 resulting binary. are there any plans to somehow "fix" that?


 LDC fixes that by using garbage collection of unused symbols by=20
 default.


=20
 Does DMD/Optlink on Windows do this too? I've noticed that using=20
 DMD/Optlink on Windows, my executable is 4MB for unittests on a=20
 module that uses a lot of CTFE template methods, while with=20
 DMD/ld on Linux (and I believe GDC/ld as well), the same test=20
 executable is over 40MB.


do `strip -s` with the executable and you will find that most of that
is various debug info.
Dec 24 2014