[dcl.meaning] General wording cleanup #3735
Conversation
source/declarations.tex
Outdated
| \tcode{T} | ||
| determines the type | ||
| \tcode{T}. | ||
| First, the type \tcode{T} is determined. |
There was a problem hiding this comment.
We have two different things we're calling T here: a decl-specifier-seq and a type. This rewrite makes it less clear that that's happening by removing the "the decl-specifier-seq T determines the type T" part. I think we should take this a bit further. How about replacing the previous paragraph ("Thus[...]") with:
Thus, a declaration of a particular declarator-id has the form attribute-specifier-seq[opt] decl-specifier-seq declarator. The optional attribute-specifier-seq appertains to the declared entity. Following is a recursive procedure for determining the type of the declared entity.
Then we only have one T, and it's a type, not a decl-specifier-seq. So we can start this second paragraph with something like:
First, a type
Tis determined from the decl-specifier-seq. [Example: ...
There was a problem hiding this comment.
Perhaps we want to talk about the type of the declarator-id or name introduced rather than the declared entity? Declarations introduce names (which denote entities), after all.
source/declarations.tex
Outdated
| and the type of the identifier in the declaration | ||
| and the type of the contained | ||
| \grammarterm{declarator-id} | ||
| in the declaration | ||
| \tcode{T} | ||
| \tcode{D1} | ||
| is ``\placeholder{derived-declarator-type-list} | ||
| \tcode{T}'', | ||
| then the type of the identifier of | ||
| then the type of the | ||
| \grammarterm{declarator-id} | ||
| in | ||
| \tcode{D} | ||
| is ``\placeholder{derived-declarator-type-list} reference to | ||
| \tcode{T}''. |
There was a problem hiding this comment.
A declarator of the form
&attribute-specifier-seq[opt] D or&&attribute-specifier-seq[opt] D determines the type of its declared entity to be the type of the entity declared by the declaration U D, where U is a type-specifier that denotes the type ''reference toT''.
(And so on for later subclauses.)
There was a problem hiding this comment.
@zygoloid The removal of derived-declarator-type-list is definitely a good idea, however, moving the form of the declarator to be inline does degrade readability, especially for longer declarator forms (i.e. functions). Perhaps we keep the form of the declarators indented as is done currently, and apply the other changes to the paragraphs?
There was a problem hiding this comment.
Also, maintaining T throughout subclauses is a little awkward, and could lead to confusion (typically, when "variables" are established, we carry them no further than then end of a paragraph, let alone a subclause)
zygoloid
added
the
changes requested
|
While I very much appreciate a general cleanup in this area, I'm mildly favoring a core issue for this, or at least CWG review of the editorial change. @zygoloid , are you reasonably sure we're not destroying |
|
@zygoloid I applied the changes you suggested, but rather than referring to "the declared entity", I changed it to refer to the declarator-id itself. |
|
@jensmaurer I walked through a few complex declarators manually following the new wording and it all checks out. |
jensmaurer
removed
the
changes requested
|
@sdkrystian: could you please rebase this (maybe after the mailing goes out), and let me know, and then I'll ask CWG to review this. The cleanup looks laudable, but I'd like some committee oversight on it. |
jensmaurer
added
the
needs rebase
|
@jensmaurer @tkoeppe I apologize for going dark for the last few years... I will be going through my open PRs and addressing comments/rebasing over the coming weeks. |
The conventions used to specify the meaning of declarators are a little inconsistent, and the sub-clause itself has a lot of single sentence paragraphs that look messy.
For example, [dcl.meaning] has a mix of specifying the "type of the identifier" and "type of the declarator-id", [dcl.ptr], [dcl.ref] and [dcl.mptr] determine the "type of the identifier", and [dcl.array] and [dcl.fct] determine the "type of the declarator-id". The mentions of "determining the type of the identifier" have been changed to determining the type of the declarator-id, as this is a better defined term.
Further, [dcl.meaning] p3 correctly acknowledged that
Tcan contain an attribute-specifier-seq, but in the very next paragraph it assumes thatTwill just be a decl-specifier-seq. [dcl.meaning] p3 is rewritten to account for that, and it additionally specified that the attribute-specifier-seq appertains to the declarator (this is also specified in [dcl.pre] p2, perhaps that mention could be removed, as it makes more sense to have it where the meaning of declarators is determined).More stylistically, [dcl.meaning] uses
codeblockwhen specifying a declaration of the forT D, which needlessly uses whitespace, and is inconsistent with the other sub-clauses. These were changed to use\tcode.Lastly, the last sentence of [dcl.meaning] p6 is completely redundant, and should be just turned into a note.