I need to parse a html document that has been generated by saving a word document as html.
I have been using the HTML agility pack quite successfully but in this instance I figured using regex for this one part might be easier (opinions?)
Word generates the following code when it translates one of its footnotes into html
<a href="#_ftn2" name="_ftnref2" title=""><span
class=MsoFootnoteReference><span class=MsoFootnoteReference><span
style='font-size:10.0pt'>[2]</span></span></span></a>
This output is consistent for every footnote with only the href= and name changing as well as the [2] text.
I need to extract the _ftn2 and [2] elements.
So far I have the following regex which will extract the _ftn2 part into the name group
<a href="#(?<name>_ftn\d).*>(<span class=MsoFootNoteReference>)
I'm having a bit of trouble parsing the second bit with all those span tags.
Is it going to be easier to use regex for this or should I continue to use the HAP for this part?
An an aside does anyone know why word generates nested identical span tags
<span class=MsoFootnoteReference>
If the input follows exactly that format then you can get away with a pretty loose regex. You just need to ignore everything except the parts you want to extract and then employ non-greedy expressions to eat up all the garbage between them:
<a href="#(?<name>_ftn\d).*?(?<number>\[\d+\]).*?<\/a>
You can use a non-greedy .*? to eat up all the extra markup because nothing in there will match your next \[\d+\] pattern. You don't really need the .*?<\/a> bit on the end, that's mostly for symmetry and a bit of extra paranoia.
Something like this is probably one of the few cases where using regular expressions to rip apart HTML makes sense. You could do this sort of thing with an HTML parser but then you'd be a nightmare of twisty XPath expressions (all of which look alike), DOM manipulations, or SAX events. And you might even get eaten by a grue.
Related
I have a bunch of records in a QuickBase table that contain a rich text field. In other words, they each contain some paragraphs of text intermingled with HTML tags like <p>, <strong>, etc.
I need to migrate the records to a new table where the corresponding field is a plain text field. For this, I would like to strip out all HTML tags and leave only the text in the field values.
For example, from the below input, I would expect to extract just a small example link to a webpage:
<p>just a small <a href="#">
example</a> link</p><p>to a webpage</p>
As I am trying to get this done quickly and without coding or using an external tool, I am constrained to using Quickbase Pipelines' Text channel tool. The way it works is that I define a regex pattern and it outputs only the bits that match the pattern.
So far I've been able to come up with this regular expression (Python-flavored as QB's backend is written in Python) that correctly does the exact opposite of what I need. I.e. it matches only the HTML tags:
/(<[^>]*>)/
In a sense, I need the negative image of this expression but have not be able to build it myself.
Your help in "negating" the above expression is most appreciated.
Assuming there are no < or > elsewhere or entity-encoded, an idea using a lookbehind.
(?:(?<=>)|^)[^<]+
See this demo at regex101
(?:(?<=>)|^) is an alternation between either ^ start of the string or looking behind for any >. From there [^<]+ matches one or more characters that are not < (negated character class).
Suppose I have HTML structured like this:
<div class="veggie">carrot</div>
<div class="veggie">cucumber</div>
<div class="fruit">
<div class="citrus">orange</div>
<div class="citrus">lemon</div>
<div class="berry">grape</div>
</div>
<div class="veggie">lettuce</div>
<div class="dairy">milk</div>
But it's all on a single line like this:
<div class="vegetable">carrot</div><div class="vegetable">cucumber</div><div class="fruit"><div class="citrus">orange</div><div class="citrus">lemon</div><div class="berry">grape</div></div><div class="vegetable">lettuce</div><div class="dairy">milk</div>
How can I translate it to XML like this:
<veggie>carrot</veggie>
<veggie>cucumber</veggie>
<fruit>
<citrus>orange</citrus>
<citrus>lemon</citrus>
<berry>grape</berry>
</fruit>
<veggie>lettuce</veggie>
<dairy>milk</dairy>
It sounds straightforward, but I have no clue where to start!
Doing this with regexes will be ugly and likely unreliable. First, regexes don't handle languages with nested structures, which HTML has. Secondly, HTML is not a clean language; it is full of errors that the browser builders in thier wisdom decided to accept, ensuring further sloppy programming by HTML writers.
A clean way to do this is to parse the HTML (just like a compiler, using a "dirty" HTML-capable parser) and build an abstract syntax tree. (You might get away with using a browser DOM). Then you apply transformations to the HTML AST to incrementally convert it into XML fragments. You can do the latter by writing ad hoc procedural code to do a recursive tree walk, check for special cases, and spit out XML. The procedureal is likely to look pretty ugly, because it is climbing up and down tree nodes, testing this, spitting that, all over the place, and the more specical cases you have, the messier this gets.
A nice way to do this is with a program transformation system (PTS). A good PTS will let you define parsers and prettyprinters for (dirty) HTML and XML; you can then parse the HTML and the PTS will will an AST as suggested in the previous paragraph. The value in the PTS is that you can usually define transformation rules using the "surface syntax" of the source and target langauges, e.g., you can say "if you see this HTML pattern, then replace it by that XML pattern. Here's a few examples:
rule replace div_class(a: attribute, t: text_content): HTMLnode -> XMLNode
= " <div class=\a>\t<div> "
-> "< \a > \t <\a /> ";
This rule matches the HTML AST (not text) for a div with only text content,
and maps it to an XML tag matching the class attribute, with the same text content, matching part of what OP wants. The double-quotes are meta quotes, to distinguish rewrite-rule syntax from source or target language syntax. The match-to part is written in HTML syntax with metavariable escapes \a and \t corresponding the values found that satisy the match. Note that this rule can only match HTML tags that contain only text as their body, because of the constraint on t. The replacement part generates the desired XML tag and content, substituting the values of the matched metavariables.
For the more complicated part of OP's example, where the HTML content is not just a text we need this rule:
rule replace div_class(a: attribute, c: content): HTMLnode -> XMLNode
= " <div class=\a>\c<div> "
-> "< \a > \c <\a /> ";
if ~ match(c,text_content);
The \c will match anything, so that's too general, but the extra "if" constraint checks that \c is NOT text_content. This rule will run where the previous rule will not, and vice-versa.
I think that covers all the OP's example, for the basics.
Without any other constraints, both rules will run whereever they can on the AST, and order of application for these rules don't matter. Conceptually, each rule converts "yellow" HTML nodes to "blue" XML nodes; collectively, the rules convert all the yellow patches to blue patches.
OP probably needs additional rules to translate the other parts of the HTML document to XML in whatever way he desires; HTML being a fairly big languages, he may have to write a bunch of rules to fill this out properly. The point is that he can write the rules largely in this same surface syntax style [as a practical matter, you often have to add some procedural code to the rules to make it all glue together right, just a lot less than the pure ad hoc way]. (Writing this as ad hoc code won't save any effort; OP will still have to handle all the HTML tag types).
Different PTS's express rules differently. I am using the rewrite rule syntax from my own PTS [DMS Software Reengineering Toolkit]. Ideally, the PTS already has available definitions of HTML and XML; DMS does.
I am following Regular Expression.info and see on their samples page an expression to match agains HTML tags, as follows:
([A-Z][A-Z0-9]*)\b[^>]*>(.*?)</\1>
What is the semantic effect of the part \b[^]? I get its a word boundary but given what follows it what is the purpose?
It matches anything extra (if it exists) up until the next occurrence of a ">" (closing HTML tag). This would capture stuff like class="classname" id="idname". However, it would also capture any character you could think of, such as •·°ÁÓ, which may or may not be what you want. As always, a proper HTML parser is the way to go for parsing HTML.
I am creating an application that will take a URL as input, retrieve the page's html content off the web and extract everything that isn't contained in a tag. In other words, the textual content of the page, as seen by the visitor to that page. That includes 'masking' out everything encapsuled in <script></script>, <style></style> and <!-- -->, since these portions contain text that is not enveloped within a tag (but is best left alone).
I have constructed this regex:
(?:<(?P<tag>script|style)[\s\S]*?</(?P=tag)>)|(?:<!--[\s\S]*?-->)|(?:<[\s\S]*?>)
It correctly selects all the content that i want to ignore, and only leaves the page's text contents. However, that means that what I want to extract won't show up in the match collection (I am using VB.Net in Visual Studio 2010).
Is there a way to "invert" the matching of a whole document like this, so that I'd get matches on all the text strings that are left out by the matching in the above regex?
So far, what I did was to add another alternative at the end, that selects "any sequence that doesn't contain < or >", which then means the leftover text. I named that last bit in a capture group, and when I iterate over the matches, I check for the presence of text in the "text" group. This works, but I was wondering if it was possible to do it all through regex and just end up with matches on the plain text.
This is supposed to work generically, without knowing any specific tags in the html. It's supposed to extract all text. Additionally, I need to preserve the original html so the page retains all its links and scripts - i only need to be able to extract the text so that I can perform searches and replacements within it, without fear of "renaming" any tags, attributes or script variables etc (so I can't just do a "replace with nothing" on all the matches I get, because even though I am then left with what I need, it's a hassle to reinsert that back into the correct places of the fully functional document).
I want to know if this is at all possible using regex (and I know about HTML Agility Pack and XPath, but don't feel like).
Any suggestions?
Update:
Here is the (regex-based) solution I ended up with: http://www.martinwardener.com/regex/, implemented in a demo web application that will show both the active regex strings along with a test engine which lets you run the parsing on any online html page, giving you parse times and extracted results (for link, url and text portions individually - as well as views where all the regex matches are highlighted in place in the complete HTML document).
what I did was to add another alternative at the end, that selects "any sequence that doesn't contain < or >", which then means the leftover text. I named that last bit in a capture group, and when I iterate over the matches, I check for the presence of text in the "text" group.
That's what one would normally do. Or even simpler, replace every match of the markup pattern with and empty string and what you've got left is the stuff you're looking for.
It kind of works, but there seems to be a string here and there that gets picked up that shouldn't be.
Well yeah, that's because your expression—and regex in general—is inadequate to parse even valid HTML, let alone the horrors that are out there on the real web. First tip to look at, if you really want to chase this futile approach: attribute values (as well as text content in general) may contain an unescaped > character.
I would like to once again suggest the benefits of HTML Agility Pack.
ETA: since you seem to want it, here's some examples of markup that looks like it'll trip up your expression.
<a href=link></a> - unquoted
<a href= link></a> - unquoted, space at front matched but then required at back
- very common URL char missing in group
- more URL chars missing in group
<a href=lïnk></a> - IRI
<a href
="link"> - newline (or tab)
<div style="background-image: url(link);"> - unquoted
<div style="background-image: url( 'link' );"> - spaced
<div style="background-image: url('link');"> - html escape
<div style="background-image: ur\l('link');"> - css escape
<div style="background-image: url('link\')link');"> - css escape
<div style="background-image: url(\
'link')"> - CSS folding
<div style="background-image: url
('link')"> - newline (or tab)
and that's just completely valid markup that won't match the right link, not any of the possible invalid markup, markup that shouldn't but does match a link, or any of the many problems with your other technique of splitting markup from text. This is the tip of the iceberg.
Regex is not reliable for retrieving textual contents of HTML documents. Regex cannot handle nested tags. Supposing a document doesn't contain any nested tag, regex still requires every tags are properly closed.
If you are using PHP, for simplicity, I strongly recommend you to use DOM (Document Object Model) to parse/extract HTML documents. DOM library usually exists in every programming language.
If you're looking to extract parts of a string not matched by a regex, you could simply replace the parts that are matched with an empty string for the same effect.
Note that the only reason this might work is because the tags you're interested in removing, <script> and <style> tags, cannot be nested.
However, it's not uncommon for one <script> tag to contain code to programmatically append another <script> tag, in which case your regex will fail. It will also fail in the case where any tag isn't properly closed.
You cannot parse HTML with regular expressions.
Parsing HTML with regular expressions leads to sadness.
I know you're just doing it for fun, but there are so many packages out there than actually do the parsing the right way, AND do it reliably, AND have been tested.
Don't go reinventing the wheel, and doing it a way that is all but guaranteed to frustrate you down the road.
OK, so here's how I'm doing it:
Using my original regex (with the added search pattern for the plain text, which happens to be any text that's left over after the tag searches are done):
(?:(?:<(?P<tag>script|style)[\s\S]*?</(?P=tag)>)|(?:<!--[\s\S]*?-->)|(?:<[\s\S]*?>))|(?P<text>[^<>]*)
Then in VB.Net:
Dim regexText As New Regex("(?:(?:<(?<tag>script|style)[\s\S]*?</\k<tag>>)|(?:<!--[\s\S]*?-->)|(?:<[\s\S]*?>))|(?<text>[^<>]*)", RegexOptions.IgnoreCase)
Dim source As String = File.ReadAllText("html.txt")
Dim evaluator As New MatchEvaluator(AddressOf MatchEvalFunction)
Dim newHtml As String = regexText.Replace(source, evaluator)
The actual replacing of text happens here:
Private Function MatchEvalFunction(ByVal match As Match) As String
Dim plainText As String = match.Groups("text").Value
If plainText IsNot Nothing AndAlso plainText <> "" Then
MatchEvalFunction = match.Value.Replace(plainText, plainText.Replace("Original word", "Replacement word"))
Else
MatchEvalFunction = match.Value
End If
End Function
Voila. newHtml now contains an exact copy of the original, except every occurrence of "Original word" in the page (as it's presented in a browser) is switched with "Replacement word", and all html and script code is preserved untouched. Of course, one could / would put in a more elaborate replacement routine, but this shows the basic principle. This is 12 lines of code, including function declaration and loading of html code etc. I'd be very interested in seeing a parallel solution, done in DOM etc for comparison (yes, I know this approach can be thrown off balance by certain occurrences of some nested tags quirks - in SCRIPT rewriting - but the damage from that will still be very limited, if any (see some of the comments above), and in general this will do the job pretty darn well).
For Your Information,
Instead of Regex, With JQuery , Its possible to extract text alone from a html markup. For that you can use the following pattern.
$("<div/>").html("#elementId").text()
You can refer this JSFIDDLE
Is there a way to indicate that two or more regex phrases can occur in any order? For instance, XML attributes can be written in any order. Say that I have the following XML:
Home
Home
How would I write a match that checks the class and title and works for both cases? I'm mainly looking for the syntax that allows me to check in any order, not just matching the class and title as I can do that. Is there any way besides just including both combinations and connecting them with a '|'?
Edit: My preference would be to do it in a single regex as I'm building it programatically and also unit testing it.
No, I believe the best way to do it with a single RE is exactly as you describe. Unfortunately, it'll get very messy when your XML can have 5 different attributes, giving you a large number of different REs to check.
On the other hand, I wouldn't be doing this with an RE at all since they're not meant to be programming languages. What's wrong with the old fashioned approach of using an XML processing library?
If you're required to use an RE, this answer probably won't help much, but I believe in using the right tools for the job.
Have you considered xpath? (where attribute order doesn't matter)
//a[#class and #title]
Will select both <a> nodes as valid matches. The only caveat being that the input must be xhtml (well formed xml).
You can create a lookahead for each of the attributes and plug them into a regex for the whole tag. For example, the regex for the tag could be
<a\b[^<>]*>
If you're using this on XML you'll probably need something more elaborate. By itself, this base regex will match a tag with zero or more attributes. Then you add a lookhead for each of the attributes you want to match:
(?=[^<>]*\s+class="link")
(?=[^<>]*\s+title="Home")
The [^<>]* lets it scan ahead for the attribute, but won't let it look beyond the closing angle bracket. Matching the leading whitespace here in the lookahead serves two purposes: it's more flexible than matching it in the base regex, and it ensure that we're matching a whole attribute name. Combining them we get:
<a\b(?=[^<>]*\s+class="link")(?=[^<>]*\s+title="Home")[^<>]+>[^<>]+</a>
Of course, I've made some simplifying assumptions for the sake of clarity. I didn't allow for whitespace around the equals signs, for single-quotes or no quotes around the attribute values, or for angle brackets in the attribute values (which I hear is legal, but I've never seen it done). Plugging those leaks (if you need to) will make the regex uglier, but won't require changes to the basic structure.
You could use named groups to pull the attributes out of the tag. Run the regex and then loop over the groups doing whatever tests that you need.
Something like this (untested, using .net regex syntax with the \w for word characters and \s for whitespace):
<a ((?<key>\w+)\s?=\s?['"](?<value>\w+)['"])+ />
The easiest way would be to write a regex that picks up the <a .... > part, and then write two more regexes to pull out the class and the title. Although you could probably do it with a single regex, it would be very complicated, and probably a lot more error prone.
With a single regex you would need something like
<a[^>]*((class="([^"]*)")|(title="([^"]*)"))?((title="([^"]*)")|(class="([^"]*)"))?[^>]*>
Which is just a first hand guess without checking to see if it's even valid. Much easier to just divide and conquer the problem.
An first ad hoc solution might be to do the following.
((class|title)="[^"]*?" *)+
This is far from perfect because it allows every attribute to occur more than once. I could imagine that this might be solveable with assertions. But if you just want to extract the attributes this might already be sufficent.
If you want to match a permutation of a set of elements, you could use a combination of back references and zero-width
negative forward matching.
Say you want to match any one of these six lines:
123-abc-456-def-789-ghi-0AB
123-abc-456-ghi-789-def-0AB
123-def-456-abc-789-ghi-0AB
123-def-456-ghi-789-abc-0AB
123-ghi-456-abc-789-def-0AB
123-ghi-456-def-789-abc-0AB
You can do this with the following regex:
/123-(abc|def|ghi)-456-(?!\1)(abc|def|ghi)-789-(?!\1|\2)(abc|def|ghi)-0AB/
The back references (\1, \2), let you refer to your previous matches, and the zero
width forward matching ((?!...) ) lets you negate a positional match, saying don't match if the
contained matches at this position. Combining the two makes sure that your match is a legit permutation
of the given elements, with each possibility only occuring once.
So, for example, in ruby:
input = <<LINES
123-abc-456-abc-789-abc-0AB
123-abc-456-abc-789-def-0AB
123-abc-456-abc-789-ghi-0AB
123-abc-456-def-789-abc-0AB
123-abc-456-def-789-def-0AB
123-abc-456-def-789-ghi-0AB
123-abc-456-ghi-789-abc-0AB
123-abc-456-ghi-789-def-0AB
123-abc-456-ghi-789-ghi-0AB
123-def-456-abc-789-abc-0AB
123-def-456-abc-789-def-0AB
123-def-456-abc-789-ghi-0AB
123-def-456-def-789-abc-0AB
123-def-456-def-789-def-0AB
123-def-456-def-789-ghi-0AB
123-def-456-ghi-789-abc-0AB
123-def-456-ghi-789-def-0AB
123-def-456-ghi-789-ghi-0AB
123-ghi-456-abc-789-abc-0AB
123-ghi-456-abc-789-def-0AB
123-ghi-456-abc-789-ghi-0AB
123-ghi-456-def-789-abc-0AB
123-ghi-456-def-789-def-0AB
123-ghi-456-def-789-ghi-0AB
123-ghi-456-ghi-789-abc-0AB
123-ghi-456-ghi-789-def-0AB
123-ghi-456-ghi-789-ghi-0AB
LINES
# outputs only the permutations
puts input.grep(/123-(abc|def|ghi)-456-(?!\1)(abc|def|ghi)-789-(?!\1|\2)(abc|def|ghi)-0AB/)
For a permutation of five elements, it would be:
/1-(abc|def|ghi|jkl|mno)-
2-(?!\1)(abc|def|ghi|jkl|mno)-
3-(?!\1|\2)(abc|def|ghi|jkl|mno)-
4-(?!\1|\2|\3)(abc|def|ghi|jkl|mno)-
5-(?!\1|\2|\3|\4)(abc|def|ghi|jkl|mno)-6/x
For your example, the regex would be
/<a href="home.php" (class="link"|title="Home") (?!\1)(class="link"|title="Home")>Home<\/a>/