Name mangling is lossy around numbers

[whitequark]

E.g. both ADC_SPC_PHASE_22_5 and ADC_SPC_PHASE_225 converge to AdcSpcPhase225.

[whitequark]

I use the following patch to the inflections crate as a workaround:

diff --git a/src/case.rs b/src/case.rs
index 0c6bea7..a22677f 100644
--- a/src/case.rs
+++ b/src/case.rs
@@ -443,6 +443,12 @@ pub fn is_constant_case(string: &str) -> bool {
   string == to_constant_case(string)
 }
 
+/// Checks if a character is a digit.
+#[inline]
+fn is_digit(c: char) -> bool {
+  c >= '0' && c <= '9'
+}
+
 /// Checks if a character is a separator.
 #[inline]
 fn is_separator(c: char) -> bool {
@@ -472,7 +478,15 @@ fn scan_to_camel(state: &mut (bool, Option<char>), curr: char) -> Option<String>
   let last = state.1;
   state.1 = Some(curr);
 
-  if state.0 {
+  if state.0 && is_digit(curr) {
+    // If the state has signaled the next character must be capitalized,
+    // but it is a digit, then prepend a separator to it.
+    state.0 = false;
+    let mut string = String::with_capacity(2);
+    string.push('_');
+    string.push(curr);
+    Some(string)
+  } else if state.0 {
     // If the state has signaled the next character must be capitalized,
     // capitalize it and mark the state as finished.
     state.0 = false;

It is not obvious how or where it should be integrated.

[whitequark]

Note: this is not perfect and results in errors such as follows:

error: variant `CcmCrcctrlInit_0` should have a camel case name such as `Ccmcrcctrlinit0`

[whitequark]

The more I think about it, the more I realize that name mangling of types (well, any name mangling that changes separators) is a bad idea.

1. It creates complications like this issue.
2. It is literally impossible to satisfy the non_camel_case_types lint without destroying semantics sometimes, because names like Foo4_20 are invalid:
   warning: type `Foo4_20` should have a camel case name such as `Foo420`
3. The specific naming of the type matters rarely because it never appears in user's code.
4. When the type name does matter is when it appears in the error messages, and I cannot in good faith describe names like SysctlSri2cR0 as in any way easy to read. It simply goes against any established practice in embedded dev.
5. It is not (visually or otherwise) symmetric with other names. E.g. uart_ibrd does not stand out as the counterpart of UartIbrd. Typical Rust code rarely has a need for having this symmetry (and rarely uses as many acronyms), but this is not typical Rust code.

Therefore I propose to allow non-camel-case types and stop this mangling altogether.

[Emilgardis]

All <name> tags are supposed to follow C ansi code standards. Code gen from SVD's usually use macro naming conventions, i.e exactly what <name> holds. This issue is a tricky one to fix.

Can one possibly make 22_5 into something that conveys the implicit .?

[whitequark]

Can one possibly make 22_5 into something that conveys the implicit .?

One could of course mangle the SVD files in any applicable way to work around a deficiency of svd2rust but I don't think this should be necessary.