Enhance mangling to handle module names

[jclark]

For subset 8, we need to be able to handle public names with

• an org that is either empty or ballerina
• arbitrary list of module names

We need to distinguish these from non-public names.

This is part of #438.

[jclark]

C++ mangling rules here: https://itanium-cxx-abi.github.io/cxx-abi/abi.html#mangling

[jclark]

Unquoted identifiers in LLVM must match [-a-zA-Z$._][-a-zA-Z$._0-9]*

https://llvm.org/docs/LangRef.html#identifiers

[jclark]

We will rely on the proposed spec restrictions here: ballerina-platform/ballerina-spec#791 (comment)

[jclark]

What we are currently doing is described here: #43 (comment)

[jclark]

GNU as syntax for names is described here: https://sourceware.org/binutils/docs/as/Symbol-Names.html#Symbol-Names

Symbol names begin with a letter or with one of ‘._’. On most machines, you can also use $ in symbol names; exceptions are noted in Machine Dependencies. That character may be followed by any string of digits, letters, dollar signs (unless otherwise noted for a particular target machine), and underscores.

[jclark]

Public name looks like:

"_B" digit+ (org-or-mod-name "__")+ local-name

The digits are a decimal number equals to (number of org-or-mod-name - 1)*4 + K, where K is

• 0 if there is a non-empty org name other than "ballerina" (so first org-or-mod-name is organization name)
• 1 if there is an empty org name (so first org-or-mod-name is a module-name)
• 2 if the org name is "ballerina" (so first org-or-mod-name is a module-name)
• 3 if the org name is "ballerina", there are two or more module names and the first module name is "lang"

So following imports would start with following prefixes

ballerina/io => _B2io__
ballerina/lang.value => _B3value__
wso2/nballerina.err => _B8wso2__nballerina__err__
foo_bar.baz => _B5foo_bar__baz__

We will keep $ for escaping individual characters.

[jclark]

Alternative scheme:

"_B" digit* letter  (org-or-mod-name "__")* local-name

In this scheme, "well-known" org/module name sequences are assigned a lower-case ASCII letter

• b ballerina/
• l ballerina/lang
• i ballerina/io
• h ballerina/http
• x ballerinax/
• w wso2/
• n (no organization) (first org-or-mod-name is module name)
• o other organization (first org-or-mod-name is org name)

The digits are a decimal number saying how many org-or-mod-name's there are. Omitting the digits is equivalent to 1.

ballerina/io => _B0i
ballerina/lang.value => _Blvalue__
wso2/nballerina.err => _B2wnballerina__err__
foo_bar.baz => _B2nfoo_bar__baz__

[jclark]

Rust mangling scheme https://rust-lang.github.io/rfcs/2603-rust-symbol-name-mangling-v0.html

[jclark]

Inspired by Rust

public-name = "_B" org-name
# "b" adds "ballerina" to start of org
org-name = ["b"] zqual mod-name
mod-name =
    nqual local-name
    | "m" nqual mod-name
local-name = ident
zqual = qual # decimal-number >= 0 
nqual = qual # decimal-number > 0
# decimal-number is the number of bytes in ident, no leading zeros
# (except zqual can be 0) 
# "_" is present if ident starts with a digit or an underscore
# must not be present if decimal-number is 0
qual = decimal-number ["_"] ident
ident = 0 or bytes

Examples

ballerina/io => _Bb02io
ballerinax/choreo => _Bb1x6choreo
ballerina/lang.value => _Bb0m4lang5value
wso2/nballerina.front.syntax=> _B4wso2m10nballerinam5front6syntax
foo_bar.baz => _B0m7foo_bar3baz

With the restrictions on org/module names ballerina-platform/ballerina-spec#791 (comment), the "_" in qual won't be needed. But we'll include this to support root module names based on filenames.

[manuranga]

@jclark What does qual stand for?

[jclark]

Qualifier. Maybe there's a better word to use here?

[manuranga]

Qualifier is alright, I just didn't think of it when I saw the abbreviation. I would have used something like part, fragment or group, no strong preference though.

[jclark]

The logic is that, in the spec, m:foo is a qualified name: the thing that the prefix m refers to is thus the qualifier of foo.

[manuranga]

Ah makes sense.