base64 decoding displays sentinel instead of incorrect character

[mzaffalon]

It is mainly commit 70c83b3. It fixes one bug and a minor problem:

• during base64 decoding, in the main loop, the error message displays the sentinel instead of the offending char
• makes base64lookup table Uint8 instead of Uint

The second commit is because the code retouching went a bit out of control.

Disclaimer: I have never done a PR, so my apologies if I am doing this wrong. And I am new to Julia too...

[mzaffalon]

By the way, I think 70c83b3 fixes also #10.

[mzaffalon]

Doh, I missed the obvious change.

[timholy]

I know @dcjones needs to take a break from using his hands, but perhaps I can help here? I'm not a Base64 coding guru, so I might need a little extra explanation. Can you give an example of the behavior that was broken before this PR, and how it now works?

If you do your development in a branch, you can easily switch back and forth. Maybe like this:

git checkout master  # make sure you are on master and have these commits
git checkout -b mz/sentinel   # Create a new branch holding these same commits. mz for your initials, sentinel for the topic.
git checkout master # to go back to master
git reset --hard origin/master  # make your master equal to the one on GitHub (deletes these commits from your master, but they're still on mz/sentinel)

Run your tests that should fail and post the results here. Then do git checkout mz/sentinel and rerun the same tests.

If at some point you want to make more commits to this branch, you can add them to mz/sentinel. Since you've already started a PR from your account's master, you can push to a different remote branch.

[timholy]

Even better would be to add your example to the tests.

Also, it looks like I will have authority to merge this. Your changes overall look fine, and might even improve performance as well.

[mzaffalon]

Thank you for the clear instructions but I think I still messed up.

Here is how the orginal code fails and how the PR fixes it:

using Codecs
valid_enc = encode(Base64, "hello world")
invalid_enc = copy(valid_enc)
invalid_enc[3] = ']'
decode(Base64, invalid_enc)

']'is not a valid Base64 character. The output of the original code is

ERROR: Invalid base64 symbol: �
 in error at error.jl:21

with the proposed change, the output is instead

ERROR: Invalid base64 symbol: ]
 in error at error.jl:21

I am not sure how to add this into a test since both the original code and the PR throw an error. It is the error message that changes.

[timholy]

Thanks for the explanation, that's very helpful.

[mzaffalon]

Thank you for taking care of it so quickly.

And because you are the one always preaching about @timeing the code, I tried, but no, I could not see any improved performance, or if there was one, it was hidden in the execution time statistical fluctuations for encoding a file of 8Mb.

Best wishes to @dcjones for a rapid recovery.