Hacker News new | ask | show | jobs
by saaaam 1489 days ago
Hi - I'd be interested to hear more details about what approaches you suggest!
4 comments

Taking the examples from https://www.youtube-nocookie.com/embed/hA1ZsxE8VJg I am sharing how I approach the simple problems in the video without using Python or having any knowledge of CSS selectors.

Retrieving the HTML

   echo https://www.nytimes.com|yy025|nc -vv proxy 80 > 1.htm
yy025 is a flexible utility I wrote to generate custom HTTP from URLs. It is controlled through environmental variables. nc is a tcpclient, such as netcat. proxy is a HOSTS file entry for a localhost TLS proxy. The sequence "yy025|tcpclient" is normally contained in a shell script that adds a <base href> tag, something like

   #! /bin/sh
   yy025 5>.1 >.2
   read x < .1;
   echo "<base href=https://$x />";
   nc -vv proxy 80 < .2|yy045;
yy045 is a utility that removes chunked transfer encoding.

The benefit of using separate, small programs that do one thing will be illustrated in the solution for Problem 3.

Problem 2 - Extract href value from <a> tags in NYT front page

Create a file called 2.l containing

    int fileno(FILE *);
    #define jmp (yy_start) = 1 + 2 *
    #define echo do {if(fwrite(yytext,(size_t)yyleng,1,yyout)){}}while(0)
   
   %s xa xb
   %option noyywrap noinput nounput
   %%
   \<a jmp xa;
   <xa>\40href=\" jmp xb;
   <xb>\" jmp 0;
   <xb>[^\"]* echo;putchar(10);
   .|\n
   %%
   int main(){ yylex();exit(0);}
Compile

    flex -8iCrf 1.l
    cc  -std=c89 -Wall -pedantic -I$HOME -pipe lex.yy.c -static -o yy1 
And finally,

    yy2 < 1.htm
This faster than Python and requires fewer resources.
It's hard to imagine an environment where the the speed/resource difference between that approach and python would matter.

Can't see reaching for something like that instead of something like

    curl -s url | htmlq a --attribute href
Problem 1 - Extract the values of <h2> tags from NYT front page

NB. In 1.htm, NYT is using the <h3> tag for headlines, not <h2> as in the 2020 video.

Solution A - Use UNIX utilties

    grep -o "<h3[^\>]*>[^\<]*" 1.htm |sed -n '/indicate-hover/s/.*\">//p'
The grep utility is ubiquitous, but the -o option is not.

https://web.archive.org/web/20201202103125/https://pubs.open...

For example, Plan9 grep does not have an -o option.

This solution is fast and flexible, but not portable.

There are myriad other portable solutions using POSIX UNIX utilities such as sh, tr and sed. For small tasks like those in "web scraping" tutorials these can still be faster than Python (due to Python start up time alone).

Solution B - Use flex to make small, fast, custom utilities

Create a file called 1.l that contains

    int fileno(FILE *);
    #define jmp (yy_start) = 1 + 2 *
    #define echo do {if(fwrite(yytext,(size_t)yyleng,1,yyout)){}}while(0)

   %s xa xb
   %option noyywrap noinput nounput
   %%
   \<h3 jmp xa;
   <xa>\> jmp xb;
   <xb>\< jmp 0;
   <xb>[^<]* echo;putchar(10);
   .|\n
   %%
   int main(){ yylex();exit(0);}
Then compile with something like

    flex -8iCrf 1.l 
    cc  -std=c89 -Wall -pedantic -I$HOME -pipe lex.yy.c -static -o yy1 
And finally,

    yy1 < 1.htm
This is faster than Python.

Solution C - Extract values from JSON instead of HTML

The file 1.htm contains a large proportion of what appears to be JSON.

I wrote a quick and dirty WIP JSON reformatter that takes web pages as input called yy059. https://news.ycombinator.com/item?id=31174088

   yy059 < 1.htm|sed -n '/promotionalHeadline\":\"[^\"]/p'|cut -d\" -f4
Sure enough, the JSON contains the headlines. One could rewrite Solution B to extract from the JSON instead of the HTML.
Problem 3 - Extract totalcount value from <span> tag in Craigslist job pages

Create a file called 3.l containing

    int fileno(FILE *);
    #define jmp (yy_start) = 1 + 2 *
   %s xa xb xc
   %option noyywrap noinput nounput
   %%
   \<ul\40id=\"jjj0\" jmp xa;
   <xa>"</ul>" yyterminate();
   <xa><a\40href=\" jmp xb;
   <xb>\" putchar(10);jmp xa;
   <xb>[^\"]* fprintf(stdout,"%s%s","https://newyork.craigslist.org",yytext);
   .|\n
   %%
   int main(){ yylex();exit(0);}
Compile

   flex -8iCrf 1.l
   cc  -std=c89 -Wall -pedantic -I$HOME -pipe lex.yy.c -static -o yy3 
yy3 extracts and prints the URLs for the job pages

Create a file called 4.l containing

    int fileno(FILE *);
    #define jmp (yy_start) = 1 + 2 *
    #define echo do{if(fwrite(yytext,(size_t)yyleng,1,yyout)){}}while(0)
   %s xa xb xc xd xe
   %option noyywrap noinput nounput
   %%
   \<h1\40class=\"cattitle\" jmp xa;
   <xa>\<a\40href jmp xb;
   <xb>\"\> jmp xc;
   <xc>[^<]* fprintf(stdout,"%s ",yytext);jmp xd;
   <xd>\<span\40class=\"totalcount\"\> jmp xe;
   <xe>\< jmp 0;
   <xe>[0-9]* echo;putchar(10);
   .|\n
   %%
   int main(){ yylex();exit(0);}
Compile

   flex -8iCrf 1.l
   cc  -std=c89 -Wall -pedantic -I$HOME -pipe lex.yy.c -static -o yy4 
yy4 extracts and prints the job catgeory name and totalcount

We can either solve this in steps where we create files or we can do it as a single pipeline. I personally find breaking a problem into discrete steps is easier.

In steps

    echo http://newyork.craigslist.org|yy025|nc -vv proxy 80|yy045 > 1.htm;
    ka;yy3 < 1.htm|yy025|nc -vv proxy 80|yy045 > 2.htm;ka-;
    yy4 < 2.htm; 
As a single pipeline

    echo http://newyork.craigslist.org|yy025|nc -vv proxy 80|y045|yy3|(ka;yy025)|nc -vv proxy 80|yy045|yy4;ka-
Shortened further by using a shell script called nc0 for the yy025|nc|yy045 sequence

    echo https://newyork.craigslist.org|nc0|yy3|(ka;nc0)|yy4
Thanks to yy025, we are using HTTP/1.1 pipelining. This is a feature of HTTP that almost 100% of httpd's support (I cannot name one that doesn't) however neither "modern" browsers nor cURL cannot take advantage of it. Multiple HTTP request are made over a single TCP connection. Unlike the Python tutorial in the video we are not "hammering" a server with multiple TCP connections at the same time, nor are we making a number of successive TCP connections that could "trigger a block". We are following the guidance of the RFCs which historically recommended that clients not open many connections to the same host at the same time. Here we only open one for retrievng all the jobs pages. Adding a delay between requests is unnecessary. We allow the server to return the results at its own pace. For most websites, this is remarkably fast. Craigslist is an anamaly and is rather slow.

What are ka and ka-. yy025 sets HTTP headers acording to environmental variables. For example, the value of Connection is set to "close" by default. To change it,

    Connection=keep-alive yy025 < url-list|nc -vv proxy 80 >0.htm
Another way is to use aliases

    alias ka="export Connection=keep-alive;set|sed -n /^Connection/p";
    alias ka-="export Connection=close;set|sed -n /^Connection/p";
    ka;yy025 < url-list|nc -vv proxy 80 >0.htm;ka-
yy025 is intended to be used with djb's envdir. Custom sets of headers can thus be defined in a directory.

This solution uses less resources, both on the client side and on the server side, than a Python approach. It is probably faster, too.