Ok, thanks for the clarification! Also, for anyone else reading this, I 
think 6.6.22.10 is the relevant part of the spec:

If the invoked function’s result type has the @noreturn attribute, 
the call expression is considered to terminate.

Does a deferred expression happen before calling into a @noreturn 
function, and if so, is this explicit in the spec? For instance, this 
program

	use bufio;
	use fmt;
	use os;
	use strings;

	export fn main() void = {
		const line = bufio::scanline(os::stdin)! as []u8;
		defer free(line);

		const line = strings::fromutf8(line);

 > /usr/bin/ld: cannot find -lc: No such file or directory

-lc is for linking to the standard C library. Something must be wrong 
with your build setup. What system are you using? On Debian you can 
just do `apt install build-essential`.

Antifennel does not mangle variables in cases that it considers simple. This
previously included all identifiers. But some identifiers need to be mangled.
For example, antifennel will convert a Lua variable called someVar to some-var
in the Fennel output. An early return that refers to some-var will need to
re-mangle this variable to make it valid in Lua. See the new test case for a
full example.
---

This revision leaves out test/fennel.lua, which had autogenerated
changes unrelated to the patch.

 anticompiler.fnl  |  4 ++--
 test.lua          |  9 +++++++++
 test_expected.fnl | 10 ++++++++++
[message trimmed]

Antifennel does not mangle variables in cases that it considers simple. This
previously included all identifiers. But some identifiers need to be mangled.
For example, antifennel will convert a Lua variable called someVar to some-var
in the Fennel output. An early return that refers to some-var will need to
re-mangle this variable to make it valid in Lua. See the new test case for a
full example.
---

Should I have committed this autogenerated change to test/fennel.lua? I think
it's only table field order changes. I can leave that out and submit a v2
patch if necessary.

 anticompiler.fnl  |  4 +--
 test.lua          |  9 +++++
[message trimmed]

On Fri, Jan 7, 2022 at 10:06 pm, Phil Hagelberg <phil@hagelb.org> wrote:

> Great catch; thanks! Applied and pushed, with a little tweak to the 
> test
> to make it a little more thorough.

That is a good addition to the test. Thanks for the review and merge!

---
 reference.md | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/reference.md b/reference.md
index cb64ef5..2dea3e0 100644
--- a/reference.md
+++ b/reference.md
@@ -1370,7 +1370,20 @@ however note that it must refer to the names after mangling has been
 done, because the identifiers must be valid Lua. The Fennel compiler
 will emit `foo-bar` as `foo_bar` in the Lua output in order for it to
 be valid. When in doubt, inspect the compiler output to see what it
-looks like.
+looks like. For example the following Fennel code:
[message trimmed]

Previously, reserved words like match would get mangled to ___match-__.
While this is technically valid, it's a really strange variable name.
The new mangling logic produces the intended ___match___.
---
 antifennel.lua          | 7 +++++--
 antifennel_expected.fnl | 7 ++++---
 test.lua                | 5 +++++
 test_expected.fnl       | 5 +++++
 4 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/antifennel.lua b/antifennel.lua
index 680473b..593bbe4 100644
--- a/antifennel.lua
+++ b/antifennel.lua
[message trimmed]

Sorry, wrong subject line. I'll resubmit this.

Previously, reserved words like match would get mangled to ___match-__.
While this is technically valid, it's a really strange variable name.
The new mangling logic produces the intended ___match___.
---
 antifennel.lua          | 7 +++++--
 antifennel_expected.fnl | 7 ++++---
 test.lua                | 5 +++++
 test_expected.fnl       | 5 +++++
 4 files changed, 19 insertions(+), 5 deletions(-)

diff --git a/antifennel.lua b/antifennel.lua
index 680473b..593bbe4 100644
--- a/antifennel.lua
+++ b/antifennel.lua
[message trimmed]