Hi,

I am getting
LNK1179: invalid or corrupt file: duplicate COMDAT '<mangled name'
The file compiles fine. What can I do?

Schobi

--
(E-Mail Removed) is never read
I'm Schobi at suespammers dot org

"Sometimes compilers are so much more reasonable than people."
Scott Meyers

Hendrik Schober <(E-Mail Removed)> wrote:
> Hi,
>
> I am getting
> LNK1179: invalid or corrupt file: duplicate COMDAT '<mangled name'
> The file compiles fine. What can I do?

Well, re-compiling the file, re-opening
the IDE, and re-booting the machine did
not help. However, commenting out the
changes I made did help and after putting
them back into place one by one it was
gone.
Scary.

Schobi

--
(E-Mail Removed) is never read
I'm Schobi at suespammers dot org

"Sometimes compilers are so much more reasonable than people."
Scott Meyers

Hendrik Schober <(E-Mail Removed)> wrote:
> Hendrik Schober <(E-Mail Removed)> wrote:
> > Hi,
> >
> > I am getting
> > LNK1179: invalid or corrupt file: duplicate COMDAT '<mangled name'
> > The file compiles fine. What can I do?
>
> Well, re-compiling the file, re-opening
> the IDE, and re-booting the machine did
> not help. However, commenting out the
> changes I made did help and after putting
> them back into place one by one it was
> gone.
> Scary.

Damn! It is back. I was afraid that
would happen.
What do I do?

Schobi

--
(E-Mail Removed) is never read
I'm Schobi at suespammers dot org

"Sometimes compilers are so much more reasonable than people."
Scott Meyers

> [...]

Anyone out there?
Whenever the file needs to get recompiled,
linking fails. Then I comment two lines of
code, build the project, and it links. Then
I go back and uncomment one of the two lines
and rebuild ==> this works. Then I go back
again, uncomment the second line and rebuild
==> it all works now.
Whenever I skip one step, it fails. Whenever
the file is recompiled (due to some header
changing), I have to start all over.
This is driving me mad.

And, of course, if I check this in, the build
might fail, too. So I'm sitting on this file
and don't know what to do!

Schobi

--
(E-Mail Removed) is never read
I'm Schobi at suespammers dot org

"Sometimes compilers are so much more reasonable than people."
Scott Meyers

Hendrik Schober wrote:
>> [...]
>
> Anyone out there?

Out here yes - haven't encountered this particular problem, nor heard of it
before, so I'm not coming up with any ideas - sorry.

> Whenever the file needs to get recompiled,
> linking fails. Then I comment two lines of
> code, build the project, and it links. Then
> I go back and uncomment one of the two lines
> and rebuild ==> this works. Then I go back
> again, uncomment the second line and rebuild
> ==> it all works now.
> Whenever I skip one step, it fails. Whenever
> the file is recompiled (due to some header
> changing), I have to start all over.
> This is driving me mad.

May you should try re-writing the code in question using a radically
different style. It seems likely that you're encountering a fairly obscure
bug in which the compiler emits two different bodies for the a single
template instantiation.

>
> And, of course, if I check this in, the build
> might fail, too. So I'm sitting on this file
> and don't know what to do!

Call Product Support.

-cd

Carl Daniel [VC++ MVP] <(E-Mail Removed)> wrote:
> [...]
> May you should try re-writing the code in question using a radically
> different style.

No way. This underlies a fairly important
part of the app. Re-designing this means
another three weeks and we are days from
a deadline.

> It seems likely that you're encountering a fairly obscure
> bug in which the compiler emits two different bodies for the a single
> template instantiation.


I don't understand this.
I suspected that it would emit two equal
bodies for different instances? But you
say, it's the other way around???

> > And, of course, if I check this in, the build
> > might fail, too. So I'm sitting on this file
> > and don't know what to do!
>
> Call Product Support.

It would take days (at least!) to sort
this out through support channels. I
am after an idea what to do today. The
problem is hard to reproduce, since
changing the source file back and forth
makes it (dis-)appear.
This puzzles me the most. Is there any
files (besides PCH's which aren't used
here) where the compiler saves any
state info between compiles???

> -cd

Schobi

--
(E-Mail Removed) is never read
I'm Schobi at suespammers dot org

"Sometimes compilers are so much more reasonable than people."
Scott Meyers

Hendrik Schober wrote:
> Carl Daniel [VC++ MVP]
> <(E-Mail Removed)> wrote:
>> [...]
>> May you should try re-writing the code in question using a radically
>> different style.
>
> No way. This underlies a fairly important
> part of the app. Re-designing this means
> another three weeks and we are days from
> a deadline.

All the more reason to start taking definitive action. Thrashing on an
obscure tool-chain bug has a low probability of success in any timeframe.

>
>> It seems likely that you're encountering a fairly
>> obscure bug in which the compiler emits two different bodies for the
>> a single template instantiation.
>
>
> I don't understand this.
> I suspected that it would emit two equal
> bodies for different instances? But you
> say, it's the other way around???

I think so. I've encountered a bug before that resulted in multiple
instantiations of a template function that were different (but have the same
name). IIUC, the linker matches up like-named COMDATs and discards all but
one - but it does check to verify that they do indeed contain the same
content.

>
>>> And, of course, if I check this in, the build
>>> might fail, too. So I'm sitting on this file
>>> and don't know what to do!
>>
>> Call Product Support.
>
> It would take days (at least!) to sort
> this out through support channels.

If you never start the process, it'll never happen. IMO it makes sense as a
contingency plan to get things rolling with PSS.

> I
> am after an idea what to do today. The
> problem is hard to reproduce, since
> changing the source file back and forth
> makes it (dis-)appear.
> This puzzles me the most. Is there any
> files (besides PCH's which aren't used
> here) where the compiler saves any
> state info between compiles???
>

Not that I know of, but there is the .ilk file where the linker saves
information if you have /incremental:yes in your link options. Normally
Release builds don't do incremental links, but that might be something to
look into.

-cd

Carl Daniel [VC++ MVP] <(E-Mail Removed)> wrote:
> [...]
> >> May you should try re-writing the code in question using a radically
> >> different style.
> >
> > No way. This underlies a fairly important
> > part of the app. Re-designing this means
> > another three weeks and we are days from
> > a deadline.
>
> All the more reason to start taking definitive action. Thrashing on an
> obscure tool-chain bug has a low probability of success in any timeframe.


I'll see if I can narrow it down tonight
as I will have to make more changes to
this file anyway.

What's interesting: It compiled fine at
first try on another machine. As long as
I can somehow make it work by doing what
I did so far, it has to remain low
priority on my todo-list until we have
passed that milestone.

> [...] I've encountered a bug before that resulted in multiple
> instantiations of a template function that were different (but have the same
> name). IIUC, the linker matches up like-named COMDATs and discards all but
> one - but it does check to verify that they do indeed contain the same
> content.

I see.

> [...]
> >> Call Product Support.
> >
> > It would take days (at least!) to sort
> > this out through support channels.
>
> If you never start the process, it'll never happen. IMO it makes sense as a
> contingency plan to get things rolling with PSS.

Mhmm. After I made it passed that deadline.

> > I
> > am after an idea what to do today. The
> > problem is hard to reproduce, since
> > changing the source file back and forth
> > makes it (dis-)appear.
> > This puzzles me the most. Is there any
> > files (besides PCH's which aren't used
> > here) where the compiler saves any
> > state info between compiles???
> >
>
> Not that I know of, but there is the .ilk file where the linker saves
> information if you have /incremental:yes in your link options. Normally
> Release builds don't do incremental links, but that might be something to
> look into.

I suppose you mean "/INCREMENTAL"?
Thanks, I will try to delete this when I
run into the problem again.

> -cd

Schobi

--
(E-Mail Removed) is never read
I'm Schobi at suespammers dot org

"Sometimes compilers are so much more reasonable than people."
Scott Meyers