To me it doesn't look like any specific imports are especially expensive, but there are a bunch of them, so they add up.
But, preliminary testing suggests that this technique works well. I imported all the invariant modules into the current namespace and stored a reference to the current namespace, as you suggested: (require mod1 mod2 ... modn) (define original-ns (current-namespace)) Then, once inside the eval namespace, attached them all: (map (λ(mod-name) (namespace-attach-module original-ns mod-name)) '(mod1 mod2 ... modn)) Now, instead of 7-8 seconds, each trip through eval takes 2-4 seconds. Probably more can be done, but that's already much better. Matthew Butterick On Mon, Oct 14, 2013 at 4:24 PM, Robby Findler <ro...@eecs.northwestern.edu>wrote: > > > > On Mon, Oct 14, 2013 at 6:20 PM, Matthew Butterick <mb.list.a...@gmail.com > > wrote: > >> I reviewed the docs about the module >> registry<http://docs.racket-lang.org/reference/syntax-model.html?q=module%20registry#%28tech._module._registry%29>, >> but just so I understand the principle behind this technique -- >> >> When you call (namespace-attach-module orig-ns 'module-name), what >> you're saying to the new namespace is "if you need module-name, use the one >> that's already been imported into orig-ns". Is this right? >> >> > Right. > > >> And if so, do I improve performance by adding another >> (namespace-attach-module >> ...) declaration for each module that will never need to be reloaded in the >> eval namespace? >> >> > Right. > > >> BTW I tried your suggestion on the quasicode, and it was indeed faster. >> However, when I moved it back into the real code, there was no improvement >> -- in fact, the real code became slower. Running the real code through >> profile suggests that most of the time is being spent traversing the >> imports (which is consistent with my observation that the complexity of the >> page itself doesn't count for much). So whatever I can do to avoid >> re-importing modules will probably be the winning technique. >> >> > Which module's imports, do you know? > > Robby > > >> Thanks. >> >> >> On Mon, Oct 14, 2013 at 11:22 AM, Robby Findler < >> ro...@eecs.northwestern.edu> wrote: >> >>> Did you try using namespace-attach-module to attach 'racket and >>> 'web-server/templates? Something like: >>> >>> (define orig-ns (current-namespace)) >>> (parameterize ([current-namespace (make-base-empty-namespace)]) >>> (namespace-attach-module orig-ns 'racket) >>> (namespace-attach-module orig-ns 'web-server/templates) >>> (namespace-require 'racket) >>> (eval `(begin >>> (require web-server/templates) >>> (require ,my-source-path-string) >>> (include-template ,my-template-path-string)) >>> (current-namespace))) >>> >>> But 7 seconds seems remarkably long (I see a speed up of about 50% for >>> that function (with the my-source-path-string and my-template-path-string >>> lines commented out): from 250 to 170 msec). >>> >>> It may be that your .zo files or something are not up to date. So you >>> may want to stick a 'raco make' somewhere in there to avoid recompiling >>> your files every time and just recompile the ones that changed. >>> >>> Robby >>> >>> >>> >>> >>> On Mon, Oct 14, 2013 at 1:14 PM, Matthew Butterick < >>> mb.list.a...@gmail.com> wrote: >>> >>>> At RacketCon I talked about my Pollen system for making web pages. >>>> Though I don't use the Racket webserver to provide live generation of web >>>> pages, I do use it to provide a kind of browser-assisted REPL for >>>> development. As I edit the project files, I want to reload the URL in the >>>> browser and see the changes (without restarting the webserver). >>>> >>>> Moreover, since it's a development tool, it wants to be maximally >>>> dynamic. For instance, I might be editing source files containing page >>>> content, or the design templates (which might also contain Racket code), or >>>> supporting Racket libraries (e.g., CSS generators). >>>> >>>> Jay has suggested that this can be done with eval. And indeed it can. >>>> Here's the quasicode I use: >>>> >>>> (parameterize ([current-namespace (make-base-empty-namespace)]) >>>> (namespace-require 'racket) >>>> (eval `(begin >>>> (require web-server/templates) >>>> (require ,my-source-path-string) >>>> (include-template ,my-template-path-string)) >>>> (current-namespace)))) >>>> >>>> This works great, with one wrinkle: it's the slowest operation in the >>>> whole system, taking about 7-8 secs to render each page. This is not a >>>> grave hardship, and certainly faster than restarting the web server. But >>>> I'd still be curious if I'm overlooking a better approach. >>>> >>>> In general, the render time is fairly consistent regardless of what's >>>> in the page, which suggests to me that most of the expense comes from >>>> setting up the fresh namespace. I did try making a namespace separately and >>>> reusing it, but that didn't work (code wouldn't get reloaded). Is there a >>>> way to "clean" a namespace that's cheaper than setting up a new one? >>>> >>>> Another option I considered is starting up another instance of Racket >>>> in a subprocess, but it would need to be made into a synchronous operation. >>>> Long drive for a short day at the beach, etc. >>>> >>>> It's also possible that I'm running up against the irreducible cost of >>>> high-quality eval, in which case, I will accept it and move on. >>>> >>>> >>>> Matthew Butterick >>>> >>>> >>>> >>>> >>>> On Thu, Jun 6, 2013 at 7:36 PM, Greg Hendershott < >>>> greghendersh...@gmail.com> wrote: >>>> >>>>> Great timing! It's been on my short list to make my static blog >>>>> generator more flexible. From >>>>> http://docs.racket-lang.org/web-server/templates.html I got the idea >>>>> that it would be suitable only for static templates, but now I >>>>> understand that's not the case. Today I made a topic branch using this >>>>> approach that's working well, which I plan to merge after some testing >>>>> and polishing. >>>>> >>>>> p.s. Maybe it's obvious, but because `include-template` uses >>>>> `include-at/relative-to/reader`, it won't include a template if you >>>>> give it a full path. Maybe there's a more-elegant way, but I found it >>>>> worked to parameterize `current-directory` around the call to `eval`. >>>>> >>>>> >>>>> On Wed, Jun 5, 2013 at 4:45 PM, Jay McCarthy <jay.mccar...@gmail.com> >>>>> wrote: >>>>> > While it is supposed to be pithy, it is also serious. >>>>> > >>>>> > When I wrote it, I imagine that people would do something like this: >>>>> > >>>>> > #lang racket/base >>>>> > (require web-server/templates) >>>>> > >>>>> > (define (template-content t x) >>>>> > (eval #`(let ([x #,x]) (include-template #,t)))) >>>>> > >>>>> > (template-content "t.txt" 5) >>>>> > (template-content "t.txt" 6) >>>>> > >>>>> > and trust that the template wouldn't include @(system "rm -fr /"). I >>>>> > think that a simple eval works just fine for when you, the >>>>> programmer, >>>>> > want to change the content dynamic (although I'd say it is better to >>>>> > use the web-server language so you can restart with the same >>>>> > continuations, etc.) >>>>> > >>>>> > Your solution is great for a template that a user provides, although >>>>> > it has the hole that the template could call @include-template and >>>>> > maybe find itself and go into a infinite loop, so it's not totally >>>>> > "secure" unless you use a sandbox. >>>>> > >>>>> > Jay >>>>> > >>>>> > On Wed, Jun 5, 2013 at 1:53 PM, Joe Gibbs Politz <j...@cs.brown.edu> >>>>> wrote: >>>>> >> I'm writing a web server in a #lang that doesn't support macros or >>>>> >> syntactic abstraction. I would like to be able to use some sort of >>>>> >> templating dynamically with bindings given by values, e.g. >>>>> >> >>>>> >> template.render("page-with-hole-for-username.html", { username: >>>>> "Joe" }) >>>>> >> >>>>> >> The FAQ ( >>>>> http://docs.racket-lang.org/web-server/faq.html#(part._.How_do_.I_use_templates__dynamically__) >>>>> ) >>>>> >> in the documentation for templates ends with this pithy >>>>> >> recommendation: >>>>> >> >>>>> >> "If you insist on dynamicism, there is always eval." >>>>> >> >>>>> >> I assume this isn't intended seriously, and instead to discourage >>>>> what >>>>> >> I'm asking about, but I'm unfortunately not seeing another way to >>>>> >> write my program. The code at the end of this message seems like >>>>> the >>>>> >> best I can do while being relatively safe, but also like a dubious >>>>> >> inclusion on a web server given the eval. >>>>> >> >>>>> >> Is this the best I can hope for? Is there another collection I >>>>> should >>>>> >> be looking into? Any other recommendations? >>>>> >> >>>>> >> Thanks! >>>>> >> Joe P. >>>>> >> >>>>> >> >>>>> >> #lang racket/base >>>>> >> >>>>> >> (require >>>>> >> web-server/templates) >>>>> >> >>>>> >> (define (render-template filename dict) >>>>> >> (define namespace-for-template (make-empty-namespace)) >>>>> >> (namespace-attach-module (current-namespace) 'web-server/templates >>>>> >> namespace-for-template) >>>>> >> (hash-map dict >>>>> >> (lambda (key value) >>>>> >> (define name-of-identifier (string->symbol key)) >>>>> >> (namespace-set-variable-value! >>>>> >> name-of-identifier >>>>> >> value >>>>> >> #f >>>>> >> namespace-for-template))) >>>>> >> (parameterize [(current-namespace namespace-for-template)] >>>>> >> (namespace-require 'web-server/templates)) >>>>> >> (define to-eval #`(include-template #,(datum->syntax >>>>> >> #'render-template filename))) >>>>> >> (eval to-eval namespace-for-template)) >>>>> >> ____________________ >>>>> >> Racket Users list: >>>>> >> http://lists.racket-lang.org/users >>>>> > >>>>> > >>>>> > >>>>> > -- >>>>> > Jay McCarthy <j...@cs.byu.edu> >>>>> > Assistant Professor / Brigham Young University >>>>> > http://faculty.cs.byu.edu/~jay >>>>> > >>>>> > "The glory of God is Intelligence" - D&C 93 >>>>> > ____________________ >>>>> > Racket Users list: >>>>> > http://lists.racket-lang.org/users >>>>> ____________________ >>>>> Racket Users list: >>>>> http://lists.racket-lang.org/users >>>>> >>>> >>>> >>>> ____________________ >>>> Racket Users list: >>>> http://lists.racket-lang.org/users >>>> >>>> >>> >> >> ____________________ >> Racket Users list: >> http://lists.racket-lang.org/users >> >> >
____________________ Racket Users list: http://lists.racket-lang.org/users
