UnicodeDecodeError: 'utf-8' codec can't decode byte 0xff in position 12192: invalid start byte

[hereischen]

When I was calling self.serial_console.get_output(), I met UnicodeDecodeError: 'utf-8' codec can't decode byte 0xff in position 12192: invalid start byte.

One possible quick solution will be:
aexpect/client.py +319

@@ -314,8 +315,8 @@  def get_output(self):
         Return the STDOUT and STDERR output of the process so far.
         """
         try:
-            with open(self.output_filename, 'r') as output_file:
-                return output_file.read()
+            with open(self.output_filename, 'rb') as output_file:
+                return output_file.read().decode(errors="ignore")
         except IOError:
             return None

What is your opinion?
@clebergnu @ldoktor @chunfuwen

[chunfuwen]

You may refer to avocado-framework/avocado-vt#1518,which use locale.getpreferredencoding() to resolve codec decode issue. Chunfu Wen

…

----- 原始邮件 ----- 发件人: "Haotong Chen" <notifications@github.com> 收件人: "avocado-framework/aexpect" <aexpect@noreply.github.com> 抄送: "chunfu wen" <chwen@redhat.com>, "Mention" <mention@noreply.github.com> 发送时间: 星期二, 2018年 5 月 15日 下午 6:08:51 主题: [avocado-framework/aexpect] UnicodeDecodeError: 'utf-8' codec can't decode byte 0xff in position 12192: invalid start byte (#31) When I am calling self.serial_console.get_output(), I met UnicodeDecodeError: 'utf-8' codec can't decode byte 0xff in position 12192: invalid start byte. One possible sulotion will be: aexpect/client.py +319 ```patch @@ -314,8 +315,8 @@ def get_output(self): Return the STDOUT and STDERR output of the process so far. """ try: - with open(self.output_filename, 'r') as output_file: - return output_file.read() + with open(self.output_filename, 'rb') as output_file: + return output_file.read().decode(errors="ignore") except IOError: return None ``` What is your opinion? @clebergnu @ldoktor @chunfuwen

-- You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub: #31

[ldoktor]

Well this looks like a different issue. According to traceback it already uses "utf-8". Anyway even "utf-8" doesn't support all kinds of combinations and \xff is reserved and can't be used by default.

To compile with py3 the get_output should return bytes, therefor the rb mode is correct. We might also consider adding get_output_text which is becoming a standard. Anyway this change will affect all places where we use it (on py3, py2 doesn't care much about bytes vs. strings).

Anyway I'll take another look at the code as more changes like this should be performed.

[ldoktor]

OK I looked at it and as it requires changes to "astring" Avocado library (due to "get_stripped_output") let's prioritize it in the next sprint. https://trello.com/c/MRIDgGi7/1322-aexpect-return-bytes-in-getoutput-and-support-for-text-version-as-well

[pevogam]

What is the progress on this issue? Can this issue be considered as a general "make aexpect sessions and similar return raw byte output similarly to avocado's process module" or do you think I should open a new issue regarding this problem? As far as I can tell not enforcing decoding of the output of various command execution functionality would solve this entirely and also be more compatible with providing reasonable outputs for commands returning pure byte output (not necessarily strings).

[ldoktor]

It looks like this card got buried under other higher-priority tasks. Anyway the basic idea of the feature is there so volunteers are welcome. @clebergnu can we move this task a bit higher?

[pevogam]

Still, if I misunderstand and this task is not relevant, I could open a new issue specifically about extending aexpect to returning bytes as cmd_output that users can decode according to their own use cases.

[ldoktor]

Yes, definitely. If cmd_output_raw (or such) is what you're looking for than that can be easily added (PR shouldn't take longer than writing an issue... 😉 )

[pevogam]

@ldoktor This will be no doubt true if adding a single method with 'rb' flags like cmd_output_raw would be sufficient. However, the entire class hierarchy is performing encoding/decoding according to an encoding attribute which is set in their constructors. What we would like for instance is to have something like

session.cmd_output("cmd-with-output-encoding-that-only-the-caller-knows-of")

and this requires fixing the entire chain of calls from the ShellSession object up to the base object. Could we maybe avoid encoding/decoding and simply deal with bytes if encoding is set to None? I guess this could break things for older code calls though.

Also, making the default output bytes and adding newer _text methods rather than keeping the current behavior and adding _raw methods is more compatible with the way it is done in other avocado repos. What do you think?

[ldoktor]

aexpect issue

[ldoktor]

@pevogam the problem is compatibility and time. I added it to our trello so we'll discuss it next week. It'd be nice to have you there (but we can follow-up on email/here)

[pevogam]

@ldoktor No problem, I will be there just in case.

[ldoktor]

Haven't seen you there (on Monday meeting), anyway we updated the card description https://trello.com/c/fqmMWhXs/1491-aexpect-enconding-issue-sync-with-avocado-usage-of-bytes and intend to sync the implementation to Avocado. That means we'll break the current behavior and Aexpect will return bytes by default while adding _text methods to return string. On py2 it shouldn't matter much, but py3 applications will have to adjust. Still the benefit of consistency across all utils seems worth the breakage.

[ldoktor]

Hello guys, as there are not many resources I created #57 that should at least easy the situation while (hopefully) not breaking anyone's workflow.

[pevogam]

Hi @ldoktor, the change is good short term solution but I hope it does not decrease the priority of the actual request here which also includes API compatibility to avocado(-vt). It also doesn't handle our case since we have a wrapper utility that would best make use of actual bytes and is not expected to modify the aexpect output in any way (read: simply pass the bytes data around without touching/converting it).

[ldoktor]

Hello @pevogam, this hot-fix is because arm's firmware puts non-utf8 chars in the output which breaks migration.with_reboot tests. We do want to fix aexpect, at this point I just don't know when as we are currently short of 2 people.