[Implemented] Structure Enhancement Request

[akj]

Could structures please be enhanced to use a boundary-alignment type, with the new syntax:

Code: Select all

Structure[.type>] <name> [Extends <name>]

Examples:
Structure.Q mystruct would start on at least a 64-bit boundary
Structure.L mystruct would start on at least a 32-bit boundary
Structure.I mystruct would start on at least a 64-bit boundary on 64-bit CPUs and at least a 32-bit boundary otherwise.

There would be an additional feature: the end of the structure would automatically be padded with nulls so that it also had [at least] the same boundary alignment as the start of the structure.

Thus

Code: Select all

Structure.q mystruct
  length.w
  height.w
  depth.a
endStructure

would start on [at least] a 64-bit boundary and be padded at the end with 24 bits to ensure that also lay on [at least] a 64-bit boundary.

The absence of a <type> would default to the current PB4.40 boundaries and if the Extends keyword was present, <type> would be ignored for the start of the structure, but honoured for the end of the structure.

[Trond]

What would be the purpose of this?

[Demivec]

Could structures please be enhanced to use a boundary-alignment type, with the new syntax:

Code: Select all

Structure[.type>] <name> [Extends <name>]

Examples:
Structure.Q mystruct would start on at least a 64-bit boundary
Structure.L mystruct would start on at least a 32-bit boundary
Structure.I mystruct would start on at least a 64-bit boundary on 64-bit CPUs and at least a 32-bit boundary otherwise.

This portion of the request would be ineffective because the structure defines the access to data, not the placement of data. For example, all structures start at offset zero, thus they are already boundary aligned. It's up to the program to determine where the structure starts.

In other words, if you defined your nice 64-bit boundary aligned structure FFIRST definition what would happen if this code was used *ptr.FFIRST = 11. How would the structure definition help you at all with your boundary alignments? This is another way of identifying it as a coding problem, and not a language problem.

There would be an additional feature: the end of the structure would automatically be padded with nulls so that it also had [at least] the same boundary alignment as the start of the structure.

Thus

Code: Select all

Structure.q mystruct
  length.w
  height.w
  depth.a
endStructure

would start on [at least] a 64-bit boundary and be padded at the end with 24 bits to ensure that also lay on [at least] a 64-bit boundary.

The absence of a <type> would default to the current PB4.40 boundaries and if the Extends keyword was present, <type> would be ignored for the start of the structure, but honoured for the end of the structure.

This portion of the request (about ending boundaries) is simply not needed as well as being ineffective (see the first example using FFIRST). If padding is needed, then add the padding.

Code: Select all

Structure mystruct ;.q
  length.w   ;2 bytes
  height.w   ;2 bytes
  depth.a    ;1 byte
  pad.b[59]  ;59 bytes
endStructure

[Fred]

In fact many API structures needs corrects 4/8 bytes alignment on 32 bits and even more on 64 bits. We wanted to add some facility for this as we spend quite some time to do it right in the residents. Wait'n'see !

[akj]

@Fred:

Thank you. That is most encouraging. So perhaps something will get implemented.