If you add a relativeToAddressIndex, then instead of making it relative to the top-most parent, you can make it relative to this object, and -1 works here as it can indicate unknown. This also allows two addresses (only if the relativeToAddressIndex
are identical and not -1) to be compared, and allows relativeness to other containers than the top most which might be more relevant. Below is an example. Empty values are unknown.
name: STACK1
kind: stack
length: -300000
offsetFromParent:
parentIndex:
relativeAddress:
relativeToAddressIndex :
absoluteAddress: 1250000
name: STACK1.frame4
kind: stackFrame
length: -2000
offsetFromParent: -1000
parentIndex: 0
relativeAddress : -1000
relativeToAddressIndex : 0
absoluteAddress: 1249000
name: variable444
kind: intData
length: 8
offsetFromParent: 32
parentIndex: 1
relativeAddress : -968
relativeToAddressIndex : 0
absoluteAddress: 1249030
On 4/29/19 3:20 PM, Larry Golding (Myriad Consulting Inc) wrote:
There s a problem with
relativeAddress , though. Here s what I currently have:
3.32.4 Relative address calculation
Each
address object has an associated value called its relative address which is
the offset of the address from the address of the top-most object in its parent chain
3.32.7 relativeAddress property
If
parentIndex ( 3.32.13 )
is present, an address object
MAY contain a property named relativeAddress whose value is an integer
containing the relative address (see 3.32.4 ) of
thisObject .
If
parentIndex is absent, then
relativeAddress SHALL be absent.
If
relativeAddress is absent, it
SHALL default to -1, which is otherwise not a valid value for this property .
That s wrong. If your top-most address is a downward-growing stack,
relativeAddress can be negative. And unlike
offsetFromParent , we can t say (I mean, we
can , but don t think we should) that relativeAddress
SHALL be present if parentIndex is present, so we can t use 0 as the default.
Ideas?
I m going to have lunch now. Back in a while.
Larry
From:
sarif@lists.oasis-open.org <
sarif@lists.oasis-open.org>
On Behalf Of Larry Golding (Myriad Consulting Inc)
Sent: Monday, April 29, 2019 12:59 PM
To: Michael Fanning
<
Michael.Fanning@microsoft.com> ; James Kupsch
<
kupsch@cs.wisc.edu> ; OASIS SARIF TC Discussion List
<
sarif@lists.oasis-open.org>
Subject: RE: [sarif] #401: Improved design of address object design
I approve this message! Much better than my solution.
From: Michael Fanning <
Michael.Fanning@microsoft.com >
Sent: Monday, April 29, 2019 12:56 PM
To: James Kupsch <
kupsch@cs.wisc.edu >; Larry Golding (Myriad Consulting Inc) <
v-lgold@microsoft.com >; OASIS SARIF TC Discussion
List <
sarif@lists.oasis-open.org >
Subject: RE: [sarif] #401: Improved design of address object design
Here s my suggestion:
For address.length, use a sentinel value of 0 meaning not provided . This makes sense, as Jim noted, a 0 length memory address isn t sensible. You don t posit describing an insertion point in memory that
shoves everything to the right. ð
For address.offsetFromParent, this property should only be consulted in cases where address.parentIndex >= 0. If address.parentIndex is -1, you ignore the offsetFromParent value (in all cases). If address.parentIndex
>= 0, the address.offsetFromParent must be populated (and it is fine if its value is 0).
This requires producers to always provided offsetFromParent when using the address parenting chain. IOW, you can t describe your parenting relationship and depend strictly on relativeAddress and/or absoluteAddress
to render those final computed values. I think this trade-off is ok, particularly as compared to consulting a separate Boolean in order to know whether you re going backward in memory.
Michael
From: James Kupsch <
kupsch@cs.wisc.edu >
Sent: Monday, April 29, 2019 12:46 PM
To: Larry Golding (Myriad Consulting Inc) <
v-lgold@microsoft.com >; Michael Fanning <
Michael.Fanning@microsoft.com >;
OASIS SARIF TC Discussion List <
sarif@lists.oasis-open.org >
Subject: Re: [sarif] #401: Improved design of address object design
The unknown is only for this address to its parents. The same problem exists whether you use -1 or unknownOffset/Length. It is useful to know offsets up to the point of where it is unknown as you at least know it is relative to this address (grand..)parent
object. The cleaner approach to working around the SDK not supporting nullable is to use the extra boolean.
On 4/29/19 2:18 PM, Larry Golding (Myriad Consulting Inc) wrote:
I really think my previous suggestion of
growthDirection is the least bad alternative if we decide NOT to use
Nullable<int> . I d modify my proposal to a
Boolean growsTowardLowMemory , default
false .
From:
sarif@lists.oasis-open.org <
sarif@lists.oasis-open.org> On Behalf Of Larry Golding (Myriad Consulting Inc)
Sent: Monday, April 29, 2019 11:46 AM
To: James Kupsch
<
kupsch@cs.wisc.edu> ; Michael Fanning
<
Michael.Fanning@microsoft.com> ; OASIS SARIF TC Discussion List
<
sarif@lists.oasis-open.org>
Subject: RE: [sarif] #401: Improved design of address object design
Importance: High
I m not as happy as I was a few moments ago. If the unknown properties default to
false , then if you don t populate
offsetFromParent (which happens at the top of every parent chain) or
length (which can happen anywhere along the parent chain), then you have to remember to set the corresponding unknown property. You won t. Of if the unknown properties default to
true , you have to set them to
false whenever you do populate offsetFromParent or
length . You won t.
Michael (addressing this question specifically to him because of the SDK impact), I don t see a good alternative to
Nullable<int> . Please let us know your opinion. Raising to red-bang.
Larry
From: Larry Golding (Myriad Consulting Inc)
Sent: Monday, April 29, 2019 11:37 AM
To: James Kupsch <
kupsch@cs.wisc.edu >; Michael Fanning <
Michael.Fanning@microsoft.com >; OASIS SARIF TC Discussion
List <
sarif@lists.oasis-open.org >
Subject: RE: [sarif] #401: Improved design of address object design
I could accept Boolean unknownLength and unknownOffsetFromParent, default false for both.
I will write it this way and send a draft. I can always change this aspect of the design if the TC objects.
Larry