cmd/gc: compiler must not accept p-notation floats in regular Go code

[griesemer]

gc accepts the p-notation for floating point numbers:

mantissa "p" exponent

such as 12p-2, etc.

http://play.golang.org/p/xOixorUxUD

The spec doesn't mention them. gccgo and go/types don't support them.

a) we should either disallow them, or
b) we should support them with consistent syntax and document them.

[griesemer]

Comment 1:

I can go either way but I would be ok supporting them.
If we do, I propose one of these productions:
pfloat_lit = int_lit ( "p" | "P" ) [ "+" | "-" ] decimals .
or
pfloat_lit = decimal_lit ( "p" | "P" ) [ "+" | "-" ] decimals .
i.e., if we permit non-decimal mantissa, we should permit any integer notation.

[robpike]

Comment 2:

Why does gc accept this? Is it needed for export data only?

[griesemer]

Comment 3:

It's plausible that it's an export data relict - I don't know of std lib code that needs
this.

[rsc]

Comment 4:

It's for export data. It's not supposed to be generally available.

[rsc]

Comment 5:

Will fix for Go 1.5. It's been there basically forever; no harm in leaving it in for Go
1.4.

[rsc]

Comment 6:

Labels changed: removed documentation, languagechange.

[griesemer]

There are several related errors:

http://play.golang.org/p/VVK5tF24e2

• p exponent
• hex mantissa and p exponent
• hex mantissa and e exponent interpreted as int

[dr2chase]

What is this "export data"? I've been trying to figure out the conditions under which 'p' and 'P' etc should not be rejected?

[minux]

The export data is what compiler read from already compiled Go packages. They are mostly regular Go declarations with some extensions (hex float, argument escape tagging, etc) Open a Go .a file under pkg directory with a text editor and you will see.

[griesemer]

"Export data" is the textual (go-syntax like) representation of the package "objects" that are exported from the package. The export data of a package A is parsed when another package B imports A. The export data can be found in the object/archive file.

The p-format is used to read and write floating-point numbers quickly and accurately without the need to convert to a decimal exponent.

[josharian]

@dr2chase see also the go/internal/gcimporter package and cmd/compile/internal/gc/export.go

[gopherbot]

CL https://golang.org/cl/10450 mentions this issue.