Of lynx and curl

I use zsh, and some aspects of this article may be zsh specific, particularly the substitution trick. bash has similar ways to achieve these goals, but I won’t be going into anything bash-specific here.

At work, I was recently tasked with archiving several thousand records from a soon-to-be-mercifully-destroyed Lotus Notes database. Why they didn’t simply ask the DBA to do this is beyond me (just kidding, it almost certainly has to do with my time being less valuable, results be damned). No mind, however, as the puzzle was a welcome one, as was the opportunity to exercise my Unix (well, cygwin in this case) chops a bit. The exercise became a simple one once I realized the database had a web server available to me, and that copies of the individual record web views would suffice. A simple pairing of lynx and curl easily got me what I needed, and I realized that I use these two in tandem quite often. Here’s the breakdown:

There are two basic steps to this process: use lynx to generate a list of links, and use curl to download them. There are other means of doing this, particularly when multiple depths need to be spidered. I like the control and safety afforded to me by this two-step process, however, so for situations where it works, it tends to be my go-to. To start, lynx --dump '' will print out a clean, human-readable version of my homepage, with a list of all the links at the bottom, formatted like


…and so on (note to self: those ./ URLs function fine, and web browsers seem to transparently ignore them, but… maybe fix that?). For our purposes, we don’t want the formatted page, nor do we want the reference numbers. awk helps us here: lynx --dump '' | awk '/http/{print $2}' looks for lines containing ‘http’, and only prints the second element in the line (default field separator being a space).

…et cetera. For my purposes, I was able to single out only the links to records in my database by matching a second pattern. If we only wanted to return links to my ‘categories’ pages, we could do lynx --dump '' | awk '/http/&&/categories/{print $2}', using a boolean AND to match both patterns.

…and so on. Belaboring this any further would be more a primer on awk than anything, but it is necessary1 for turning lynx --dump into a viable list of URLs. While this seems like a clumsy first step, it’s part of the reason I like this two-step approach: my list of URLs is a very real thing that can be reviewed, modified, filtered, &c. before curl ever downloads a byte. All of the above examples print to stdout, so something more like lynx --dump '' | awk '/http/&&/categories/{print $2}' >> categories-urls would (appending to and not clobbering) store my URLs in a file. Then it’s on to curl. for i in $(< categories-urls); curl -O "$i" worked just fine2 for my database capture, but our example here would be less than ideal because of the pretty URLs. curl will, in fact, return

curl: Remote file name has no length!

…and stop right there. This is because the -O option simplifies things by saving the local copy of the file with the remote file’s name. If we want to (or need to) name the files ourselves, we use the lowercase -o filename instead. While this would be a great place to learn more about awk3, we can actually cheat a bit here and let the shell help us. zsh has a tail-matching substitution built in, used much like basename to get the tail end of a path. Since URLs are just paths, we can do the same thing here. To test this, we can for i in $(< categories-urls); echo ${i:t}.html and get


…blah, blah, blah. This seems to work, so all we need to do is plug it in to our curl command, for i in $(< categories-urls); (curl -o "${i:t}".html "$i"; sleep 2). I added the two seconds of sleep when I did my db crawl so that I wasn’t hammering the aging server. I doubt it would have made a difference so long as I wasn’t making all of these requests in parallel, but I had other things to work on while it did its thing anyway.

One more reason I like this approach to grabbing URLs – as we’re pulling things, we can very easily sort out the failed requests using curl -f, which returns a nonzero exit status upon failure. We can use this in tandem with the shell’s boolean OR to build a new list of URLs that have failed: (i=""; curl -fo "${i:t}".html "$i" || echo "$i" >> failed-category-urls) gives us…

  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
  0     0    0     0    0     0      0      0 --:--:-- --:--:-- --:--:--     0
curl: (22) The requested URL returned error: 404 Not Found
~% < fail.html
zsh: no such file or directory: fail.html
zsh: exit 1      < fail.html
~% < failed-category-urls

…which we can then run through curl again, if we’d like, to get the resulting status codes of these URLs: for i in $(< failed-category-urls); (printf "$i", >> failed-category-status-codes.csv; curl -o /dev/null --location --silent --head --write-out '%{http_code}\n' "$i" >> failed-category-status-codes.csv)4. < failed-category-status-codes.csv in this case gives us,404

…which we’re free to do what we want with. Which, in this case, is probably nothing. But it’s a good one-liner anyway.

  1. sed and/or grep could sub in here, but awk is really the right tool for this one. ↩︎
  2. Well, mostly. The filenames were all things like 0bbc93e72b0c16d7852580d8004ef57e?OpenDocument, which I tidied up a tiny bit using zmv: zmv '(*)?OpenDocument' '$1.html' ↩︎
  3. This would be something like awk -F/ '$(NF-1)', with -F specifying the field separator, $NF being a variable that represents the number of fields in a line, and then backtracking by one because of the trailing slash. ↩︎
  4. There’s a bit to unpack here, and it’s beyond the scope of this article, so this is where I refer the reader to man curl. Suffice it to say, however, I use this curl one-liner to check status codes rather often. ↩︎