Add multi-label support to Find box for number fields

[rvisser7]

I've had a go at implementing issue #6882 in this PR, just for number fields for now. It essentially follows the SneakyBox approach suggested by @roed314 .

As always, any comments/feedback are very welcome! 🙂 At present, this is only implemented for number fields, but I'll hopefully extend this to other sections soon.

How it works:

• A function multi_entry_jump_search has been added to search_wrapper.py. This is meant to be a generic handler which parses a comma-separated list of entries (e.g. could be labels/names/polynomials/equations) given as an input string in the jump box. Each entry is processed using a custom section-specific parser function parse_entry. (e.g. for number fields, this is just nf_string_to_label).

  If any of the entries are unable to be parsed, it flashes an info message on the number of invalid entries. If all entries are invalid, it flashes an error message and returns the usual home search page.

• Also in search_wrapper.py, a function parse_labels has been added to convert a "?labels=..." URL query into a database query of the form {"$in": [...]} on the label column.

• I've tried to keep the section-specific code to a minimum. In particular, the only update to number_field.py is some code at the beginning of number_field_jump which first runs the generic multi_entry_jump_search parser, and setting the label_knowl argument (used to define the labels SneakyBox).

Examples for testing:

• If there's no comma in the jump box, then it treats the input as usual. E.g.

  http://localhost:37777/NumberField/?jump=2.2.5.1
  http://localhost:37777/NumberField/?jump=Qsqrt7
  http://localhost:37777/NumberField/?jump=x%5E2+-+3

• E.g: a given list of labels:

  http://localhost:37777/NumberField/?jump=2.2.5.1%2C+2.2.8.1%2C+2.2.12.1

• E.g. a mix of labels, nicknames, polynomials:

  http://localhost:37777/NumberField/?jump=2.2.5.1%2C+Qsqrt11%2C+x%5E3+-+2%2C+Qzeta10

• E.g. a mix of labels and invalid entries:

  http://localhost:37777/NumberField/?jump=3.3.49.1%2C+banana%2C+x%5E2+-+7

• E.g. some random nonsense:

  http://localhost:37777/NumberField/?jump=1234%2C+banana%2C+%21%21%21%2C+asdfghjk

For now, I've just put this as a draft, just to get any preliminary feedback on whether this looks ok, or whether this should maybe be implemented in a different way. If the editors are happy with the above implementation, I can then implement this for all other sections of the LMFDB where we'd like to support a multi-label search in the "Find" jump box. :)

[roed314]

I think you could just do labels = list(set(label.strip() for label in labels_input.split(","))).

[rvisser7]

thanks, done!

[rvisser7]

Perhaps this doesn't matter too much, but would it maybe make more sense for parse_labels to instead be in the utils/search_parsing.py file?

[roed314]

@jwj61 expressed concern about polredabs getting called many times. I wonder if you could also add a progressive timeout here, where you stop if the total amount of parsing time surpasses an amount determined by a keyword.

[rvisser7]

Ah, of course, thanks - I forgot about this!

I've added a timer to this for loop. If the timer hits the value set by time_limit (I've put a default of 30 seconds), then it flashes an error and returns the index page. At the moment, the timer is only checked between each entry being parsed, so I am assuming that parsing at least a single entry won't take too long.

[roed314]

Maybe you can provide sep as a function, defaulting to lambda x: re.split(",", x). Then you can allow Q(sqrt2,sqrt3) for a field name even though it has a comma in it by making a more complicated splitting function.

[rvisser7]

Thanks, this is a great suggestion!

I've had a go at writing a function split_top_level_commas, currently placed just above multi_entry_jump_search. This takes some input string and returns a list of substrings which only splits on commas which are not inside any parentheses/brackets/braces. Since this would probably give the intended behaviour for most sections (not just number fields), I've made this the default separator function.

[roed314]

technically this is the input to url_for, not a url itself.

[rvisser7]

thanks, fixed!

[rvisser7]

Just for fun, maybe I can also mention another nice consequence of this PR: this also provides a convenient way get a search page for essentially any parametrised family of fields directly via the Find box (or the Labels box). So in particular, I think this also at least gives some partial progress towards issue #6948 🙂

E.g. to obtain a search results page for the first few cyclotomic fields (ordered by degree), we can just copy-paste the Python output of ",".join("Qzeta"+str(n) for n in range(100)) into the Find box. Or similarly, to get a search for pure cubic fields, we just paste the output of ",".join("x^3-"+str(n) for n in range(1000)) .

Just for convenience, I've given some links to search pages for some of the families mentioned in #6948 below. To estimate the polredabs cost, I've also given some rough estimates on the time taken for each page to load using the "jump" link, (measured using the Legendre server):

• Cyclotomic fields $\mathbb{Q}(\zeta_n)$, for $n \leq 100$ (jump, labels). Takes < 1 sec to load.

• Maximal totally real subfields $\mathbb{Q}(\zeta_n)^+$ of cyclotomics, for $n \leq 100$ (jump, labels). Takes < 1 sec to load.

• Pure cubic fields $\mathbb{Q}(\sqrt[3]{a})$ for $0 \leq a &lt; 1000$ (jump, labels). Takes ~11 sec to load.

• Pure quartic fields $\mathbb{Q}(\sqrt[4]{a})$ for $0 \leq a &lt; 1000$ (jump, labels). Takes ~12 sec to load.

• Simplest cubic fields $\mathbb{Q}[x]/(x^3 - ax^2 - (a+3)x - 1)$ for $0 \leq a &lt; 1000$ (jump, labels). Takes ~11 sec to load.

• Ennola's cubic fields $\mathbb{Q}[x]/(x^3 + (a-1)x^2 - ax - 1)$ for $0 \leq a &lt; 1000$: (jump, labels). Takes ~10 sec to load.