Re: [pbs-devel] [PATCH v2 pxar 01/20] decoder/aio: add contents() and content_size() calls

[Stefan Reiter <s.reiter@proxmox.com>]

On 3/25/21 10:08 AM, Wolfgang Bumiller wrote:
> On Wed, Mar 24, 2021 at 04:18:08PM +0100, Stefan Reiter wrote:
>> Returns a decoder::Contents without a wrapper type, since in this case
>> we don't want to hide the SeqRead implementation (as done in
>> decoder::sync). For conviencience also implement AsyncRead if "tokio-io"
>> is enabled.
>>
>> Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
>> ---
>>
>> proxmox-backup requires a dependency bump on this!
>>
>> v2:
>> * make contents() call available without tokio-io feature
>> * drop peek() implementation
>>
>>   src/decoder/aio.rs | 36 +++++++++++++++++++++++++++++++++++-
>>   1 file changed, 35 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/decoder/aio.rs b/src/decoder/aio.rs
>> index 82030b0..ee4d51f 100644
>> --- a/src/decoder/aio.rs
>> +++ b/src/decoder/aio.rs
>> @@ -5,7 +5,7 @@ use std::io;
>>   #[cfg(feature = "tokio-fs")]
>>   use std::path::Path;
>>   
>> -use crate::decoder::{self, SeqRead};
>> +use crate::decoder::{self, Contents, SeqRead};
>>   use crate::Entry;
>>   
>>   /// Asynchronous `pxar` decoder.
>> @@ -56,6 +56,16 @@ impl<T: SeqRead> Decoder<T> {
>>           self.inner.next_do().await.transpose()
>>       }
>>   
>> +    /// Get a reader for the contents of the current entry, if the entry has contents.
>> +    pub fn contents(&mut self) -> Option<Contents<T>> {
>> +        self.inner.content_reader()
>> +    }
>> +
>> +    /// Get the size of the current contents, if the entry has contents.
>> +    pub fn content_size(&self) -> Option<u64> {
>> +        self.inner.content_size()
>> +    }
>> +
>>       /// Include goodbye tables in iteration.
>>       pub fn enable_goodbye_entries(&mut self, on: bool) {
>>           self.inner.with_goodbye_tables = on;
>> @@ -67,6 +77,7 @@ mod tok {
>>       use std::io;
>>       use std::pin::Pin;
>>       use std::task::{Context, Poll};
>> +    use crate::decoder::{Contents, SeqRead};
>>   
>>       /// Read adapter for `futures::io::AsyncRead`
>>       pub struct TokioReader<T> {
>> @@ -93,6 +104,29 @@ mod tok {
>>               }
>>           }
>>       }
>> +
>> +    impl<'a, T: crate::decoder::SeqRead> tokio::io::AsyncRead for Contents<'a, T> {
>> +        fn poll_read(
>> +            self: Pin<&mut Self>,
>> +            cx: &mut Context<'_>,
>> +            buf: &mut tokio::io::ReadBuf<'_>,
>> +        ) -> Poll<io::Result<()>> {
>> +            unsafe {
>> +                // Safety: poll_seq_read will *probably* only write to the buffer, so we don't
>> +                // initialize it first, instead we treat is a &[u8] immediately and uphold the
>> +                // ReadBuf invariants in the conditional below.
>> +                let write_buf =
>> +                    &mut *(buf.unfilled_mut() as *mut [std::mem::MaybeUninit<u8>] as *mut [u8]);
>> +                let result = self.poll_seq_read(cx, write_buf);
>> +                if let Poll::Ready(Ok(n)) = result {
>> +                    // if we've written data, advance both initialized and filled bytes cursor
>> +                    buf.assume_init(buf.filled().len() + n);
>> +                    buf.advance(n);
> 
> The differences here look fishy. The docs about `assume_init()` say it
> marks the "first `n` *unfilled* bytes`, so IMO both calls should use
> just `n` as they both start from the beginning of the unfilled range?
> 

Hm, you're right: 
https://docs.rs/tokio/1.4.0/src/tokio/io/read_buf.rs.html#240

That doc is a bit confusing IMO...

>> +                }
>> +                result.map(|_| Ok(()))
>> +            }
>> +        }
>> +    }
>>   }
>>   
>>   #[cfg(feature = "tokio-io")]
>> -- 
>> 2.20.1