Repository navigation
Conversation
Codecov Report
@@ Coverage Diff @@
## master #2643 +/- ##
==========================================
+ Coverage 93.14% 93.16% +0.02%
==========================================
Files 61 61
Lines 12130 12166 +36
==========================================
+ Hits 11298 11335 +37
+ Misses 832 831 -1
Continue to review full report at Codecov.
|
st-pasha
left a comment
st-pasha
left a comment
There was a problem hiding this comment.
Interesting... So basically on any kind of error you're saying "What if I tried with a different kind of quoting rule instead, would I be able to parse then?" Then usually you won't, but on rare occasions it does work, which means the initial guess of the QR was incorrect and it was fixed. And since we don't care that much about speed for invalid inputs, there's virtually no cost to do this extra check. Great!
I guess the only remaining inconsistency is that half of the input was parsed under one QR, while the other half under a different QR. Say, if first half of the file had QR=1 (doubled quotes), while the second half had QR=2 (escaped quotes), then it won't be flagged as an error. Admittedly, we never saw that happening IRL, so maybe nothing to worry about...
|
👍 But I like how you expressed this in terms of just another re-read pass -- I thought it would be much harder than that... |
|
Yes exactly. If a quote rule type bump occurs there is a warning on that line always, so at least the user knows something might be improper in the file. Some refinement needed for sure but it's erring on the side of caution for now (always warns if any quote rule bump occurred). Hm. That's not true if there's a one-row footer auto removed too, currently. Will improve that ... |
…tests those lines according to gdb
4 participants
Closes #2265
test()no longer relies onoption(warn=2). Multiple warnings() are now tested, as needed for the new test for this issue.\ninoutput=is now left in rather than stripped out. Oroutput=can be a vector of lines, too.