Adding /TEST to the linker switch might be useful, so when you do see this
quirky IL behavior it may give you a hint, or you can mail it to us and see
what we can derive from it. We may need to request a link repro though to
get to the bottom of it.
Thanks,
Kang Su Gatlin
Visual C++ Program Manager
--------------------
| From: "Steve Baer" <
[email protected]>
| References: <O#
[email protected]>
<
[email protected]>
| Subject: Re: Kang Su Gatlin: Linker Whitepaper question
| Date: Tue, 10 Aug 2004 09:05:12 -0700
|
| Thanks for the quick reply and for helping me understand your point.
|
| Our application exposes a large API for 3rd party developers to write
| plug-ins, so we are stuck on this issue.
|
| We've had some quirky issues with the incremental linker in the past, and
| are trying to find ways to fix this. Our main issue is that the IL will
work
| just fine for a while and all of a sudden completely fail. This forces us
to
| do a full rebuild. I don't have the error code on me, but I will pass
along
| the word to everyone in the company to write down the error message when
the
| IL fails.
|
| Thanks again,
| -Steve Baer
| Robert McNeel & Associates
|
|
| | > Yes, I can see the confusion that can arise from that statement.
| > __declspec(export) in itself is not a bad thing. The code maintenance
| > problem has to do with exporting functions that you don't need to
export,
| > as now you have exports that later developers added to the code base are
| > wondering "who else uses this??".
| >
| > Let me rewrite the snippet and see if this is more clear:
| > "The linker's handling of directives left something to be desired in
| > VC2003, and for this reason code with lots of directives can suffer. In
| > particular __declspec(export), as it's likely the most common linker
| > directive. Given this, don't export functions that you don't need to
| > export. Exposing exports that are not necessary can be confusing to
| > developers working on the codebase, and it additionally increases your
| link
| > times. Of course creating exports for your APIs with
__declspec(export)
| > is exactly what you should do. For VS2005 we do plan on fixing some of
| > these speed issues related to directives. I'll update this note when
that
| > happens (although the recommendation to export only what you want to
| expose
| > stays intact regardless)."
| >
| > I hope this makes things a bit more clear. If not, feel free to let me
| > know. Either on here or directly at (e-mail address removed).
| >
| > I'll post an updated version of the note shortly.
| >
| > Thanks,
| >
| > Kang Su Gatlin
| > Visual C++ Program Manager
| >
| > --------------------
| > | From: "Steve Baer" <
[email protected]>
| > | Subject: Kang Su Gatlin: Linker Whitepaper question
| > | Date: Mon, 9 Aug 2004 23:36:43 -0700
| >
| > |
| > | I recently read your whitepaper under the "extremely long link times"
| post
| > | and have a question that I was hoping you could answer.
| > |
| > | My question is based on the following paragraph:
| > | Directives
| > | The linker's handling of directives left something to be desired in
| > VC2003,
| > | and for this reason code with lots of directives can suffer. In
| > particular
| > | __declspec(export), as it's likely the most common linker directive.
| > | Besides that, don't export functions that you don't need to export.
| > | Besides it becoming a code maintenance nightmare, it also increases
your
| > | link times.
| > |
| > |
| > | Is there some other "better" way to handle DLL exports that makes the
| > | "maintenance nightmare" more manageable?
| > | Or do you just think that creating APIs is a bad idea?
| > |
| > | Thanks for writing this article and I hope you can help clear up my
| > | questions.
| > | -Steve
| > |
| > |
| > |
| >
|
|
|
