> The syntax doesn't seem noticeably clearer?

From the very first example:

    csvgrep -c 1 -m ILLINOIS 

using coreutils grep you'd need something like this:

    grep -E '^ILLINOIS,' 2009.csv 

Oh, wait, that doesn't work because the file was last saved by something which quotes it.

    grep -E '^"?ILLINOIS"?,'

except that would produce unexpected results with 'ILLINOIS"' so it really needs to be something like:

    grep -E '^(["]{0,1})ILLINOIS\1,'

Bear in mind that this is the simplest possible case and doesn't even touch on issues like quoting in the shell or needing to handle files which have embedded separators as data values (imagine what happens when our erstwhile grep data-miner needs to check the stats for `WASHINGTON, DISTRICT OF COLUMBIA`…).

In all but the most trivial cases it's safer and easier to use tools designed for the job. csvkit also has the very nice property that it's callable as a Python library so when you outgrow a simple shell processing pipeline you could migrate your notes / scripts to a full Python program without having to retest everything related to basic file I/O, unicode, etc. which you would otherwise need to do when switching readers.

(Bear in mind that the author works in data journalism – the target user is not a grizzled Unix expert but someone who has a bunch of CSV files and a full time job which is not shell scripting)

> And - without testing - I presume csvkit in Python is a bit slower than the GNU coreutils in C?

Perhaps but it'd be unlikely for it to be noticeable for n less than millions on a remotely modern system – the Python regular expression engine has quite decent performance and if that became an issue, PyPy will even JIT them for you. In the very few cases I've seen where there is a noticeable difference it was always because the Python version was decoding Unicode and the shell tools were running with LC_ALL=C, which meant that corrupt data made it further before being caught and, in some cases, either failing to match all of the records or subtly quietly things by not extracting all of a combined character, etc.

For the target use-case, however, this is likely all to be many, many orders of magnitude less than the time most people would spend debugging regex patterns.

I take you're point and those other well made points about line continuations below -- but you're over egging it here as

cut -d, -f1 2010.csv | grep ILLINOIS

works just fine for this data. I emphasise again - I take your overall point though.

Yeah, I'm more sensitive to this class of errors now that I live in DC since the "Washington, District of Columbia" format is common enough to show up sporadically.

I also don't really understand why you want to grep the whole block instead of the single word 'ILLINOIS'.

My examples anchored it to avoid matching outside of the expected field. That probably wouldn't matter for the easy data used in the csvkit example but, for example, suppose you were looking at business data and your search for California included records from every state with a California Pizza Kitchen, California Tortilla, etc. Or, the reverse, you want your search for Washington to include records from the state but not DC.

This class of error is somewhat treacherous since it's common for people not to notice it before they start working with a full data file. Using tools which don't require constant caution to avoid data errors is simply a basic good habit.

Grep understands text files, not CSV syntax. They're not equivalent. For example, the following CSV document has two lines, but grep treats them as three:

  1,2,5,"Hello"
  3,4,5,"How are
  you doing?"

That's correct, and as you illustrate it's the possibility to have newlines and commas inside quoted fields that complicates things for grep/awk/cut/etc.

So instead of making a more complex version of tools like grep, we can make the data simple for these tools to understand. That's what https://github.com/dbro/csvquote does. It can be run in a pipeline before the grep stage, and allow grep/cut/awk/... to work with unambiguous field and record delimiters. Then it can restore the newlines and commas inside the quoted fields at the end of the pipeline.

They're not replacements for those tools. They're new tools that work with a specific tabular data format.

The data I come across is a mix of comma, tab, and space delimited though predominantly what the commenter above would call "line oriented".

So I personally just don't see the benefit from a less generalised version of coreutil tools

... but fair does I'm sure there are plenty who have use for a more specific tool.

> So I personally just don't see the benefit from a less generalised version of coreutil tools

Others have pointed out specific examples, but the general reason (which I don't think I've seen explicitly articulated) is that CSV isn't strictly line oriented -- for intsance, records can span multiple "physical" lines, because line breaks within quoted strings are part of the field value, not separate records -- so line-oriented tools don't work with CSV generally (they may if you happen to have CSV where nothing inconsistent with a line-oriented view happens to be done, but that's luck not something that is generally true of CSV.)

I'm an experienced Unix admin, and I actually wrote my own versions of these csv tools long ago instead of using coreutils. I had a lot of different datasets to with with, and it turned it there were so many inconsistencies in the CSV data that I needed a tool that could handle it all.

It also became super simple to extract and munge data with options for tabular data. With two tiny commands I can exclude rows, include rows, sed columns, merge tables, and export new tables, all with just csv files. So it's pretty handy.

And don't forget AWK!

I've been playing with these old school tools lately on medium sized tabular data files and I've been impressed by the performance and capabilities.