"Anthra Norell" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
>
> Paul,
>
> I think self-respect is a very essential attitude and I am pleased to know
that you value it as much as I do.
> The off topic thus dispatched, let me first express my appreciation
or your interest. Next I agree to anything you say about
> pyparsing, because I know little about it. Perhaps you will agree that
parsing is overkill if you don't need to. Correct me if I'm
> wrong, but I have a hunch that HTM tags don't nest. So you can actually
strip them with two or three lines of plain python code,
> looking at the letters one by one.
> As to my proposal I can only invite you to run my program first and
then pass judgment, unlike the clerical Authorities in the
> story who steadfastly refused to take a peek through Galileo's telescope,
arguing that, since planets didn't have rings, taking a
> peek was an exercise of utter futility.
> And do tell me what a TOS is. If I am missing a point it is rather
this one. Does it have to do with copyright? How would it
> be then if you did your pyparsing demo on a site with a more accommodating
TOS?
>
> Regards
>
> Frederic
>
>
Frederic -
HTML parsing is one of those slippery slopes - or perhaps "tar babies" might
be a better metaphor - that starts out as a simple problem, but then one
exception after the next drags the solution out for daaaays. Probably once
or twice a week, there is a posting here from someone trying to extract data
from a website, usually something like trying to pull the href's out of some
random <a> (is for "anchor") tags, usually followed by a well-meaning regexp
suggestion along the lines of "Use '<a href="[^"]*">'", followed by "oh
wait, I forgot to mention that sometimes there are other attributes in the
tag besides href, or sometimes the referenced url is not in double quotes,
but single quotes, or no quotes, or href is sometimes HREF, or Href, or
there are spaces around the '=' sign, or spaces before the closing ">", or
there are two spaces between the 'a' and 'href', followed by "no problem,
just use *this* regexp: (@#\$(#&$)q(\/\/\<*w&)(Q&!!!!". You are correct -
in general, HTML tags themselves do not nest, and this is certainly a
blessing. But HTML tags expressions *do* nest - lists within lists, tables
within tables, bold within italic, italic within bold, etc. And HTML, being
less well-behaved than XML, also allows tags to overlap, as in
<foo><bar></foo></bar>. Not to mention, some HTML contains comments, so
href's inside comments *probably* want to be ignored.
So utilities like BeautifulSoup, targeted specifically at cracking
less-than-pristine HTML, come to the fore, to bring some order to this
parsing chaos. pyparsing is more generic than BeautifulSoup, and I believe
less HTML-capable, especially with overlapping tags. But it does include
some out-of-the-box helpers, such as makeHTMLTags("tagname"), which
generates opening and closing <tagname> and </tagname> tag expressions,
including handling for odd casing, unpredictable whitespace, and
unanticipated tag attributes. And for applications such as this, pyparsing
(I believe) provides some programming interfaces that are easier to deal
with than the BeautifulSoup accessors I've seen posted here.
I'm less and less open to the "parsing is overkill, I'll just slap together
a regexp" line of reasoning. With a little practice, pyparsing apps can
come together fairly quickly, and keep their readability for a long time. I
find regexp's, which I *don't* use every day, and whose typoglyphics quickly
bleed out of my frontal lobe, to lose their readability over time,
especially when the expression to be cracked contains any of the regex
special characters, namely (,),[,],.,/,^,$,+,*,?,\,... they become a riddle
of backslashes. The difficulty is that regexp's commingle their semantic
punctuation with the text they are trying to process, and it demands an
extra mental focus to decipher them (doubly so when trying to debug them!).
I try to save my (limited number of remaining) mentally focused moments for
things like overall system design, and so I wrote pyparsing to help separate
the syntax and mental processing used for text content vs. pattern
symbology. Pyparsing uses explicit Python classes, such as Optional,
OneOrMore, Group, Literal, and Word, with *some* operator shortcuts, such as
+ for And, | for match-first-alternative Or, ^ for match-longest-alternative
Or, and ~ for Not. Yes, the expressions are not as terse as regexps, and
they are not as fast at runtime. But they are quickly graspable at a
glance. Here is a BNF for part of the Verilog language, and the
corresponding code in the Verilog language parser:
<UDP> ::= primitive <name_of_UDP> ( <name_of_variable>
<,<name_of_variable>>* ) ;
<UDP_declaration>+
<UDP_initial_statement>?
<table_definition>
endprimitive
udp = Group( "primitive" + identifier +
"(" + Group( delimitedList( identifier ) ) + ")" + semi +
OneOrMore( udpDeclaration ) +
Optional( udpInitialStmt ) +
udpTableDefn +
"endprimitive" )
and the udp expression is inherently tolerant of random embedded whitespace
(and comments too, as provided for later in the code).
Is it overkill to use this same tool to allow me to write:
anchorStart,anchorEnd = makeHTMLTags("a")
print [ t.href for t,s,e in anchorStart.scanString(htmlSourceText) ]
to list out the href's in an arbitrary body of HTML, independent of
whitespace, presence of other attributes, casing of tag and attribute names,
and other typical pitfalls of HTML parsing? (Note: this 2-liner does not
handle HTML comments, this would require at least one more line, something
like "aStart.ignore(htmlComment)", but I haven't tested this for this
mini-manifesto.)
Frederic, I'm sorry, but I probably wont get around to giving SE a try. I'm
not that interested in becoming proficient in writing (or even just being
able to read) expressions such as these:
Tag_Stripper = SE.SE ('"~<.*?>~= " "~<[^>]*~=" "~[^<]*>~=" ')
CSV_Maker = SE.SE (' "~\s+~=(9)" ')
CSV_Maker = SE.SE ('"~\s+~=,"')
If these make sense to you, more power to you. If I were to puzzle them out
this evening, the gift would be lost within a week. I can't afford to lose
my grasp of code I've written in so short a time.
As for "TOS", it is "Terms of Service," and it's kind of like copyright.
It's one of those unilateral contracts that basically says, "I offer this
information, but only if you follow my terms - such as, don't rip off my
data just to set up a competing service, or to sell the data in some
repackaged form - and if you don't like my terms, you're welcome to not use
my service" (liberally translated from the legalese). Friendly consumer
sites like Google, UPS, FedEx, Amazon all list some form of TOS stating
"you're welcome to use our website as long as it is within the bounds of
normal consumer use - and no bots, agents, etc. allowed, 'cause they'll
bring our server to its knees and the innocent bystander users end up losing
out because of your selfishness/greed/lack of consideration for your fellow
human person." As much as the content on the Internet is freely available,
it isn't always free. (Pyparsing comes with some simple HTML scrapers that
use public websites, such as NIST's listing of public NTP servers.)
So what started out as a little joke (microscopic, even) has eventually
touched a nerve, so thanks and apologies to those who have read this whole
mess. Frederic, SE looks like a killer - may it become the next regexp!
-- Paul
--
http://mail.python.org/mailman/listinfo/python-list
