[nem-bug] [Nemerle 0000701]: [0.9.3] Empty structs have
$PLACE_HOLDER$ field
feedback at nemerle.org
feedback at nemerle.org
Sun Jul 2 15:52:26 CEST 2006
A NOTE has been added to this issue.
======================================================================
<http://nemerle.org/bugs/view.php?id=701>
======================================================================
Reported By: Snaury
Assigned To: nazgul
======================================================================
Project: Nemerle
Issue ID: 701
Category: Compiler
Reproducibility: always
Severity: feature
Priority: normal
Status: feedback
======================================================================
Date Submitted: 07-01-2006 16:05 CEST
Last Modified: 07-02-2006 15:52 CEST
======================================================================
Summary: [0.9.3] Empty structs have $PLACE_HOLDER$ field
Description:
The following code:
public struct EmptyStruct { }
produces empty structure with the following IL:
.class public sequential ansi sealed beforefieldinit EmptyStruct
extends [mscorlib]System.ValueType
{
.pack 0
.size 1
} // end of class EmptyStruct
Nemerle's generated structure has no the following IL:
.class public auto ansi sealed EmptyStruct
extends [mscorlib]System.ValueType
{
.field private specialname uint8 $PLACE_HOLDER$
} // end of class EmptyStruct
As you can see, they differ and Nemerle's IL has strange $PLACE_HOLDER$
field. If possible, it would be good if generated code would be without
this $PLACE_HOLDER$ field...
======================================================================
----------------------------------------------------------------------
nazgul - 07-01-06 21:26
----------------------------------------------------------------------
AFAIR as added this field to mimick the C# compiler. But maybe has changed
after .NET 1.1 or it was mono compiler's behaviour. It must be checked if
this is still needed.
----------------------------------------------------------------------
nazgul - 07-02-06 12:56
----------------------------------------------------------------------
After using a patch I get verification error for empty structs:
bash-3.00$ ../../tools/verify out.exe
Microsoft (R) .NET Framework PE Verifier. Version 2.0.50727.42
Copyright (c) Microsoft Corporation. All rights reserved.
[MD]: Error: ClassLayout has parent TypeDef token=0x02000002 marked
AutoLayout. [token:0x00000001]
1 Error Verifying out.exe
I'm investigating the possible fix.
----------------------------------------------------------------------
Snaury - 07-02-06 13:11
----------------------------------------------------------------------
Seems like after that class needs to be marked Sequential. I don't yet know
how type lookup in compiler works, though, so I don't know how a
StructLayout custom attribute can be applied on that struct...
----------------------------------------------------------------------
nazgul - 07-02-06 13:15
----------------------------------------------------------------------
At the stage of emitting hierarchy by S.R.E we can just define this
attribute using code similar to make_nemerle_variant_attribute in
HierarchyEmitter.n
----------------------------------------------------------------------
nazgul - 07-02-06 13:46
----------------------------------------------------------------------
I enchanced the patch and commited it, thanks!
Fixed on trunk (r6416).
----------------------------------------------------------------------
Snaury - 07-02-06 15:28
----------------------------------------------------------------------
Unless it's not too late, by occasion I found that you don't need to apply
attribute to achieve sequential layout, plus I found that both mcs and c#
default to sequential layout for structs (which can later be overriden by
applying attribute in user code). Besides, I found a way to better mimick
mcs behaviour, in regard for example to beforefieldinit flag (set
beforefieldinit when there is no static constructor). I'll attach new
patch against r6417.
----------------------------------------------------------------------
nazgul - 07-02-06 15:41
----------------------------------------------------------------------
Shouldn't it be Static in the GetConstructor call? Do you have the exact
reference to specification, when beforefieldinit is emitted?
----------------------------------------------------------------------
Snaury - 07-02-06 15:52
----------------------------------------------------------------------
Actually yes, it should have been static (typo), but for some reason static
doesn't work there and throws (I'm trying to figure out why). :-/ I don't
have specification, but looking at mcs sources I see that they do it when
there is no default constructor, ms c# *seems* to do the same, I don't
know though. I think it's not even about specification, it's about
optimization, because if there is no static constructor -> there is no
initialization of fields -> we shouldn't request runtime to initialize
class upon first static call.
Issue History
Date Modified Username Field Change
======================================================================
07-01-06 16:05 Snaury New Issue
07-01-06 20:30 Snaury File Added: bug-placeholder.patch
07-01-06 21:26 nazgul Note Added: 0001331
07-02-06 12:56 nazgul Note Added: 0001334
07-02-06 13:11 Snaury Note Added: 0001335
07-02-06 13:15 nazgul Note Added: 0001336
07-02-06 13:46 nazgul Status new => resolved
07-02-06 13:46 nazgul Resolution open => fixed
07-02-06 13:46 nazgul Assigned To => nazgul
07-02-06 13:46 nazgul Note Added: 0001337
07-02-06 13:46 nazgul Description Updated
07-02-06 15:28 Snaury Status resolved => feedback
07-02-06 15:28 Snaury Resolution fixed => reopened
07-02-06 15:28 Snaury Note Added: 0001340
07-02-06 15:28 Snaury Description Updated
07-02-06 15:29 Snaury File Added: bug-placeholder-2.patch
07-02-06 15:41 nazgul Note Added: 0001341
07-02-06 15:52 Snaury Note Added: 0001342
======================================================================
More information about the bugs
mailing list