der: pem reader rejects zero lengths reads

[baloo]

Since base64ct 1.7.0 (#1387), we now rejects zero lengths reads. This happens to trigger with an sha256WithRSAEncryption AlgorithmIdentifier where the parameters are null.

[baloo]

I'm actually wondering if a rollback might not be better though. base64ct 1.7 is going to break der 0.7 too.
Either

--- a/base64ct/src/decoder.rs
+++ b/base64ct/src/decoder.rs
@@ -105,7 +106,7 @@
     /// - `Err(Error::InvalidLength)` if the exact amount of data couldn't be read, or
     ///   if the output buffer has a length of 0
     pub fn decode<'o>(&mut self, out: &'o mut [u8]) -> Result<&'o [u8], Error> {
-        if out.is_empty() {
+        if self.is_finished() && out.is_empty() {
             return Err(InvalidLength);
         }

or:

--- a/base64ct/src/decoder.rs
+++ b/base64ct/src/decoder.rs
@@ -106,7 +107,7 @@
     ///   if the output buffer has a length of 0
     pub fn decode<'o>(&mut self, out: &'o mut [u8]) -> Result<&'o [u8], Error> {
         if out.is_empty() {
-            return Err(InvalidLength);
+            return Ok(out);
         }

         if self.is_finished() {

[tarcieri]

Yeah, sounds like we should revert the change in base64ct and publish another 1.7 release

[tarcieri]

Opened #1714 to revert the change

[tarcieri]

@baloo I think we can potentially merge this as well as #1714, if you want to fix rustfmt

[baloo]

Closed in favor of #1716, the test case was merged there.