pub struct InputHandle { /* fields omitted */ }
Input handles are basically Read objects with a few extras. We don't
require the standard io::Seek because we need to provide a dummy
implementation for GZip streams, which we wouldn't be allowed to do
because both the trait and the target struct are outside of our crate.
An important role for the InputHandle struct is computing a cryptographic
digest of the input file. The driver uses this information in order to
figure out if the TeX engine needs rerunning. TeX makes our life more
difficult, though, since it has somewhat funky file access patterns. LaTeX
file opens work by opening a file and immediately closing it, which tests
whether the file exists, and then by opening it again for real. Under the
hood, XeTeX reads a couple of bytes from each file upon open to sniff its
encoding. So we can't just stream data from read()
calls into the SHA2
computer, since we end up seeking and reading redundant data.
The current system maintains some internal state that, so far, helps us Do
The Right Thing given all this. If there's a seek on the file, we give up
on our digest computation. But if there's a seek back to the file
beginning, we are open to the possibility of restarting the computation.
But if nothing is ever read from the file, we once again give up on the
computation. The ExecutionState
code then has further pieces that track
access to nonexistent files, which we treat as being equivalent to an
existing empty file for these purposes.
Consumes the object and returns the underlying readable handle that
it references.
Consumes the object and returns the SHA256 sum of the content that was
written. No digest is returned if there was ever a seek on the input
stream, since in that case the results will not be reliable. We also
return None if the stream was never read, which is another common
TeX access pattern: files are opened, immediately closed, and then
opened again.
Various piece of TeX want to use the libc ungetc()
function a lot.
It's kind of gross, but happens often enough that we provide special
support for it. Here's getc()
emulation that can return a previously
ungetc()
-ed character.
Here's the ungetc()
emulation.
Pull some bytes from this source into the specified buffer, returning how many bytes were read. Read more
🔬 This is a nightly-only experimental API. (read_initializer
)
Determines if this Read
er can work with buffers of uninitialized memory. Read more
Read all bytes until EOF in this source, placing them into buf
. Read more
Read all bytes until EOF in this source, appending them to buf
. Read more
Read the exact number of bytes required to fill buf
. Read more
Creates a "by reference" adaptor for this instance of Read
. Read more
Transforms this Read
instance to an [Iterator
] over its bytes. Read more
Deprecated since 1.27.0
: Use str::from_utf8 instead:
https://doc.rust-lang.org/nightly/std/str/struct.Utf8Error.html#examples
🔬 This is a nightly-only experimental API. (io
)
the semantics of a partial read/write of where errors happen is currently unclear and may change
Transforms this Read
instance to an [Iterator
] over [char
]s. Read more
Creates an adaptor which will chain this stream with another. Read more
Creates an adaptor which will read at most limit
bytes from it. Read more