💾 Archived View for gemi.dev › gemini-mailing-list › 000302.gmi captured on 2023-11-04 at 12:39:51. Gemini links have been rewritten to link to archived content
-=-=-=-=-=-=-
Hello all, I was excited to learn about Gemini recently. I ported some web content of mine over (gemini://pboyd.io), and I've been enjoying looking through what's out there. The Gemini sites (err.. capsules?) remind me of the web in another era, so I thought it would be fun to re-create some "classic" CGI scripts. I started with a Guestbook (gemini://pboyd.io/guestbook/, source: gemini:// pboyd.io/guestbook/guestbook.pl). I got it working, but it felt like I was in the weeds from the very start. The first problem I hit is that I'm limited to one input at a time. I wanted a few inputs (name, location, comment). Of course, plenty of command line programs work fine with a single argument list, so, in theory, that should be enough. But flags or comma-separated input wasn't what I was going for. Successive input prompts didn't seem quite right either. I ended up with a "form" made out of links to input URLs. The data is collected in a temp file named with a random session ID, and there's a final "submit" link that updates the generated guestbook (in true retro CGI fashion it probably fails horribly when two people submit at once). This works, but like I said, kinda in the weeds. Is there a better way to get multiple inputs that I missed? The other problem I ran into is that dynamic responses were cached in my client. In my case, the user would submit the form and be presented with an error page. They'd fix the error and resubmit only to get the same error again because of the client cache. I worked around that by making sure each generated link is unique (by a counter). Conditional GET behavior is probably too complicated, but it would be nice if there were a "Pragma: no-cache" type flag on dynamic pages. Anyway, am I making this harder than it needs to be? Am I just trying to do something that shouldn't be done on Gemini? Cheers, Paul
One way to do it would be to have multiple inputs in a row, and store the results in a temp file based on IP address. It still fails if multiple people from the say IP address submit, but that's much rarer. makeworld
That's a good idea. It would probably work fine for this particular case. Though I'd worry about it for anything more complicated. But I guess that could be solved by client-side certs. As an aside, I had a co-worker who made a web game once and added IP-based rate limiting. The only trouble was that the only people who played his game were his co-workers who mostly played from the office network, totally triggering his rate limit. On Fri, Jul 17, 2020 at 11:35 AM <colecmac at protonmail.com> wrote: > One way to do it would be to have multiple inputs in a row, > and store the results in a temp file based on IP address. It > still fails if multiple people from the say IP address submit, > but that's much rarer. > > makeworld >
I like the form idea and I wrote something similar for gopher a while back [0]. https://gopher.floodgap.com/gopher/gw?=mozz.us+7005+312f64656d6f2d666f726d In my case, I used URL query parameters to keep track of previously submitted form values. This worked because query params do not have any special meaning in gopher URLs. For gemini the query component is reserved, but you could probably accomplish the same thing using a special path component with the encoded fields in it. gemini://example.com/cgi-bin/form/name%3Dpaul%26location%3Dcanada/ - mozz [0] https://github.com/michael-lazar/flask-gopher/blob/master/demo/run_server.py#L154
Hmm.. interesting. I'm already packing data into the URL path, so yeah, I can just put the rest of the data in there. I'll have to try it out when I get a minute. The protocol spec has a max URL size of 1024 bytes, but as long as the inputs are short I think it will work. Paul On Fri, Jul 17, 2020 at 1:37 PM Michael Lazar <lazar.michael22 at gmail.com> wrote: > I like the form idea and I wrote something similar for gopher a while back > [0]. > > https://gopher.floodgap.com/gopher/gw?=mozz.us+7005+312f64656d6f2d666f726d > > In my case, I used URL query parameters to keep track of previously > submitted > form values. This worked because query params do not have any special > meaning > in gopher URLs. For gemini the query component is reserved, but you could > probably accomplish the same thing using a special path component with the > encoded fields in it. > > gemini://example.com/cgi-bin/form/name%3Dpaul%26location%3Dcanada/ > > - mozz > > [0] > https://github.com/michael-lazar/flask-gopher/blob/master/demo/run_server.py#L154 >
A word of warning - as you probably have surmised, query string or otherwise manipulated URLs are not good for all types of application. They work just like HTTP GET queries and as such should really be limited to idempotent interactions with the server (like search queries or other peristent URLs). There is nothing yet equivalent to a non-cacheable data submission to the server (like HTTP POST) yet. I think it would be great if there could be one, but nothing on the horizon at the moment. So the complex URLs you can assemble in interactions with Gemini are subject to being linked to and replayed. For example if I have a page on my site with 100 parametrised links to a gemini based guest book, every time GUS comes to visit, there could end up being 100 new entries posted! Best wishes - Luke On 17-Jul-2020 19:26, Paul Boyd wrote: > Hmm.. interesting. I'm already packing data into the URL path, so > yeah, I can just put the rest of the data in there. I'll have to try > it out when I get a minute. > > The protocol spec has a max URL size of 1024 bytes, but as long as the > inputs are short I think it will work. > > Paul > > On Fri, Jul 17, 2020 at 1:37 PM Michael Lazar > <lazar.michael22 at gmail.com <mailto:lazar.michael22 at gmail.com>> wrote: > > I like the form idea and I wrote something similar for gopher a > while back [0]. > > https://gopher.floodgap.com/gopher/gw?=mozz.us+7005+312f64656d6f2d666f726d > > In my case, I used URL query parameters to keep track of > previously submitted > form values. This worked because query params do not have any > special meaning > in gopher URLs. For gemini the query component is reserved, but > you could > probably accomplish the same thing using a special path component > with the > encoded fields in it. > > gemini://example.com/cgi-bin/form/name%3Dpaul%26location%3Dcanada/ > <http://example.com/cgi-bin/form/name%3Dpaul%26location%3Dcanada/> > > - mozz > > [0] > https://github.com/michael-lazar/flask-gopher/blob/master/demo/run_server.py#L154 >
I put all the collected data in the URL. It works pretty well and got rid of the temp files. It event fixed the cache problem since each generated URL is unique to the data submitted. But it generates some nasty URLs (see the bottom). Luke, good points about the ability to replay a post. Temporary session files didn't have that problem at least. I suppose I could bring back a session ID just to counter replays, but I'll wait until it becomes a problem. Anyway, I'm probably not going to write something like this again. The guestbook is kind of fun (Hello K?vin and Timur!), so I'll leave it up. But I'll stick to things that fit better in the future. Case in point about replays, here's an ugly URL to a pre-filled form: gemini:// pboyd.io/cgi-bin/guestbook.pl/link/S2luZyBBcnRodXI=;Q2FtZWxvdA==;VGhlIExhZH kgb2YgdGhlIExha2UsIGhlciBhcm0gY2xhZCBpbiB0aGUgcHVyZXN0LCBzaGltbWVyaW5nIHNhb Wl0ZSwgaGVsZCBhbG9mdCBFeGNhbGlidXIgZnJvbSB0aGUgYm9zb20gb2YgdGhlIHdhdGVyIHNp Z25pZnlpbmcgYnkgRGl2aW5lIFByb3ZpZGVuY2UgdGhhdCBJLCBBcnRodXIsIHdhcyB0byBjYXJ yeSBFeGNhbGlidXIuIFRoYXQgaXMgd2h5IEkgYW0geW91ciBraW5nIQ==;?none Paul On Fri, Jul 17, 2020 at 2:59 PM Luke Emmet <luke at marmaladefoo.com> wrote: > A word of warning - as you probably have surmised, query string or > otherwise manipulated URLs are not good for all types of application. > They work just like HTTP GET queries and as such should really be > limited to idempotent interactions with the server (like search queries > or other peristent URLs). > > There is nothing yet equivalent to a non-cacheable data submission to > the server (like HTTP POST) yet. I think it would be great if there > could be one, but nothing on the horizon at the moment. > > So the complex URLs you can assemble in interactions with Gemini are > subject to being linked to and replayed. > > For example if I have a page on my site with 100 parametrised links to a > gemini based guest book, every time GUS comes to visit, there could end > up being 100 new entries posted! > > Best wishes > > - Luke > > On 17-Jul-2020 19:26, Paul Boyd wrote: > > Hmm.. interesting. I'm already packing data into the URL path, so > > yeah, I can just put the rest of the data in there. I'll have to try > > it out when I get a minute. > > > > The protocol spec has a max URL size of 1024 bytes, but as long as the > > inputs are short I think it will work. > > > > Paul > > > > On Fri, Jul 17, 2020 at 1:37 PM Michael Lazar > > <lazar.michael22 at gmail.com <mailto:lazar.michael22 at gmail.com>> wrote: > > > > I like the form idea and I wrote something similar for gopher a > > while back [0]. > > > > > https://gopher.floodgap.com/gopher/gw?=mozz.us+7005+312f64656d6f2d666f726d > > > > In my case, I used URL query parameters to keep track of > > previously submitted > > form values. This worked because query params do not have any > > special meaning > > in gopher URLs. For gemini the query component is reserved, but > > you could > > probably accomplish the same thing using a special path component > > with the > > encoded fields in it. > > > > gemini://example.com/cgi-bin/form/name%3Dpaul%26location%3Dcanada/ > > <http://example.com/cgi-bin/form/name%3Dpaul%26location%3Dcanada/> > > > > - mozz > > > > [0] > > > https://github.com/michael-lazar/flask-gopher/blob/master/demo/run_server.py#L154 > > >
> On Jul 18, 2020, at 15:46, Paul Boyd <boyd.paul2 at gmail.com> wrote: > > I put all the collected data in the URL. It works pretty well and got rid of the temp files. It event fixed the cache problem since each generated URL is unique to the data submitted. But it generates some nasty URLs (see the bottom). Ugly indeed. Also, keep in mind that URLs are limited to 1024 bytes in Gemini.
On Saturday, July 18, 2020 3:46 PM, Paul Boyd <boyd.paul2 at gmail.com> wrote: > I put all the collected data in the URL. It works pretty well and got rid of the temp files. It event fixed the cache problem since each generated URL is unique to the data submitted. But it generates some nasty URLs (see the bottom). Consequence is heavy url, possibly bigger than the resulting server response. Session token and server side temporary file is nicer on network usage... Maybe in the future, the spec could evolve to allow efficient form, managed by the gemini client and submitted to the server without the need of multiple connection. freD.
Hi! On Sat, 18 Jul 2020 at 16:47, Paul Boyd <boyd.paul2 at gmail.com> wrote: > Anyway, I'm probably not going to write something like this again. The guestbook is kind of fun (Hello K?vin and Timur!), so I'll leave it up. But I'll stick to things that fit better in the future. I think it's really valuable to *try* all kinds of things and different approaches to see *what* fits well. IMHO we have only scratched the surface of what is possible with the Gemini protocol (and what makes for good UX). To summarize what has been discussed here: 1. Filling in multiple values is not trivial with the input response type 2. One way is to generate a session id on the server. (This is stateful, which has some downsides but avoids replay issues.) 3. Another way is to embed all the responses in the URL. Some ideas that haven't been mentioned: a. If you include all the fields at once, you could just ask for them in one request. Eg. "10 name;location;link;comment" which the user is smart enough to parse and respond with "Hannu;;;This is cool!". Or for a better UX, make the fields newline-separated (ie. %0A in the URL) and have multiline-capable input fields in browsers. Note: this is against a strict interpretation of the spec but hasn't been discussed AFAIK. b. Instead of having different links for different fields, you could ask for the fields in succession (but this only works well for the session id based approach). Eg. /guestbook/sign responds with "30 /guestbook/sign/aF3d", then /guestbook/sign/aF3d responds with "10 Name", /guestbook/sign/aF3d?Hannu with "10 Location (optional)", then "10 Link (optional)" and "10 Comment" in sequence, saving each query string to the server-side session. After the final response either save the comment or show the responses with a link to save. I personally think the Gemini protocol (as it currently is) can be as powerful as the UNIX command line (note: only CLI, not TUI). In fact, you could serve a remote shell over Gemini. Of course we're all spoiled by interactive software even in a terminal so Gemini feels quite limited. -Hannu
I did consider a succession of inputs. I think we can all agree that a succession of modal inputs is irritating, and the UI would be exactly that on most clients. Asking the user to use a separator seemed error prone, so I didn't want to do that either. I know it can work. makeworld's gemlikes has a prompt of: > Enter your username, a space, and your comment Usernames don't have spaces, and comments do, so this works great here. I'm just not sure people are going to get "name;location;link;comment" right consistently. I certainly wouldn't, I'd end up with "paul;;my comment" instead of "paul;;;my comment" and get my comment in the link field. Your UNIX command-line comment is interesting. One set of input args is enough for UNIX, so maybe getopt style flags? --name John --location "From Beyond" --comment "This is what I have to say" A form still seems right for a guestbook (the real-life version is a form after all). But there ought to be something interesting we can do with flags on the input line. Maybe an ASCII art drawing program? With inputs like "line --from 0,0 --to 10,10". Paul On Sun, Jul 19, 2020 at 5:28 AM Hannu Hartikainen < hannu.hartikainen+gemini at gmail.com> wrote: > Hi! > > On Sat, 18 Jul 2020 at 16:47, Paul Boyd <boyd.paul2 at gmail.com> wrote: > > Anyway, I'm probably not going to write something like this again. The > guestbook is kind of fun (Hello K?vin and Timur!), so I'll leave it up. But > I'll stick to things that fit better in the future. > > I think it's really valuable to *try* all kinds of things and > different approaches to see *what* fits well. IMHO we have only > scratched the surface of what is possible with the Gemini protocol > (and what makes for good UX). > > To summarize what has been discussed here: > 1. Filling in multiple values is not trivial with the input response type > 2. One way is to generate a session id on the server. (This is > stateful, which has some downsides but avoids replay issues.) > 3. Another way is to embed all the responses in the URL. > > Some ideas that haven't been mentioned: > > a. If you include all the fields at once, you could just ask for them > in one request. Eg. "10 name;location;link;comment" which the user is > smart enough to parse and respond with "Hannu;;;This is cool!". Or for > a better UX, make the fields newline-separated (ie. %0A in the URL) > and have multiline-capable input fields in browsers. Note: this is > against a strict interpretation of the spec but hasn't been discussed > AFAIK. > > b. Instead of having different links for different fields, you could > ask for the fields in succession (but this only works well for the > session id based approach). Eg. /guestbook/sign responds with "30 > /guestbook/sign/aF3d", then /guestbook/sign/aF3d responds with "10 > Name", /guestbook/sign/aF3d?Hannu with "10 Location (optional)", then > "10 Link (optional)" and "10 Comment" in sequence, saving each query > string to the server-side session. After the final response either > save the comment or show the responses with a link to save. > > I personally think the Gemini protocol (as it currently is) can be as > powerful as the UNIX command line (note: only CLI, not TUI). In fact, > you could serve a remote shell over Gemini. Of course we're all > spoiled by interactive software even in a terminal so Gemini feels > quite limited. > > -Hannu >
On Sun Jul 19, 2020 at 2:59 PM CEST, Paul Boyd wrote: > I did consider a succession of inputs. I think we can all agree that a > succession of modal inputs is irritating, and the UI would be exactly > that > on most clients. To my mind this is the canonical way of getting multi-part input: use client certificates to establish a persistent session and then just have the server send successive responses with status code 10. I agree this would be irritating in a GUI client that pops up a box akin to a HTTP basic auth login box every time, but that's more an argument against that being a good UI design than against this being a good way to handle this kind of input. This approach is what the protocol makes natural/obvious, and so it makes sense to design UIs to match it. In an UI like AV-98, this approach works exceptionally well and provides a look and feel that's almost indistinguishable from using a shell, or a terminal application with a REPL interface. That's a very powerful thing which we still haven't even begun to experiment with the power of. I realise something like AV-98 is not a comfortable "daily reading" interface for most people. Of course, there are probably ways to make this approach pleasant to use in a GUI client - people are very welcome to work on that. I also still think that having separate clients for reading static text and using remote applications makes a lot of sense. It's not necessarily a good match to quick and easy, low-effort interactivity - but then, that was never the goal. Cheers, Solderpunk
You could have a sort of landing page for the form, using client certificates, that looks something like Foo: $foo => /form/foo Enter Foo Bar: $bar => /form/bar Enter Bar => /form/submit Submit Form where it displays the current value (if it exists), each associated link allows that value to be input/changed, and then the submit link marks the whole form as ready to be processed. You could even substitute "Please fill out all fields" or something for the last link until the required information is all available. On July 19, 2020 10:01:51 AM EDT, Solderpunk <solderpunk at posteo.net> wrote: >On Sun Jul 19, 2020 at 2:59 PM CEST, Paul Boyd wrote: >> I did consider a succession of inputs. I think we can all agree that >a >> succession of modal inputs is irritating, and the UI would be exactly >> that >> on most clients. > >To my mind this is the canonical way of getting multi-part input: use >client certificates to establish a persistent session and then just >have >the server send successive responses with status code 10. > >I agree this would be irritating in a GUI client that pops up a box >akin >to a HTTP basic auth login box every time, but that's more an argument >against that being a good UI design than against this being a good way >to handle this kind of input. This approach is what the protocol makes >natural/obvious, and so it makes sense to design UIs to match it. > >In an UI like AV-98, this approach works exceptionally well and >provides >a look and feel that's almost indistinguishable from using a shell, or >a >terminal application with a REPL interface. That's a very powerful >thing which we still haven't even begun to experiment with the power >of. > >I realise something like AV-98 is not a comfortable "daily reading" >interface for most people. Of course, there are probably ways to make >this approach pleasant to use in a GUI client - people are very welcome >to work on that. I also still think that having separate clients for >reading static text and using remote applications makes a lot of sense. >It's not necessarily a good match to quick and easy, low-effort >interactivity - but then, that was never the goal. > >Cheers, >Solderpunk -- Sent from my Android device with K-9 Mail. Please excuse my brevity.
On Sun, 19 Jul 2020 at 17:08, Solderpunk <solderpunk at posteo.net> wrote:
> In an UI like AV-98, this approach works exceptionally well and provides
> a look and feel that's almost indistinguishable from using a shell, or a
> terminal application with a REPL interface. That's a very powerful
> thing which we still haven't even begun to experiment with the power of.
I was going to post something like this. The paradigm chosen by Gemini
basically prevents GUI or TUI like applications, but it's great for
console use. The web is already a universal GUI application platform,
so Gemini can be something else.
An interaction method like the one in AV-98 would be impossible for
the web but it's amazing for Gemini. I believe this is a strength
we're only beginning to understand. For example, consider the huge
rise of podcasts and audiobooks: people love to listen to information.
With Gemini it would be trivial to browse with a headset while
commuting. The web, on the other hand, is horrible for screen reader
use (ask any blind person).
On Sun, 19 Jul 2020 at 15:59, Paul Boyd <boyd.paul2 at gmail.com> wrote:
> Your UNIX command-line comment is interesting. One set of input args is
enough for UNIX, so maybe getopt style flags?
>
> --name John --location "From Beyond" --comment "This is what I have to say"
I like this idea! It also makes me wonder if single-line-capable data
formats would be useful. Eg. edn {:name "Hannu" :comment "Hello
world!"} and YAML {name: Hannu, comment: Hello world!}. (Both do
suffer from unfamiliarity, though.) There must be a huge array of
alternatives, most of them better than my CSV idea.
-Hannu
On Sun, 2020-07-19 at 17:50 +0200, mbays at sdf.org wrote: > *Except* that you can only have one line of output between each line > of > input, because 10 doesn't allow a body. This is rather limiting. > I think this can work because you use redirects in between (which is also what you need to do if you want to encode the answers in the URL). /story/123/comment/add 20 text/gemini\r Hello! We're going to ask you for a temporary certificate to establish a session, your name, and the comment, OK? If you're ready, click the link below! => ready /story/123/comment/ready 60 Please tell your client to generate a temporary certificate\r /story/123/comment/ready 20 text/gemini\r Excellent! First your name, then the comment, OK? If you're ready, follow the link below. => name /story/123/comment/name 10 Your name? \r 30 gemini://example.org/story/123/comment/name-ok /story/123/comment/name-ok 20 text/gemini\r Excellent, Alex. Thanks! Now for the comment. Please remember that you have about 900 characters. Follow the link below when you're ready. => comment 10 Let's have that comment!\r 30 gemini://example.org/story/123/comment/comment-ok /story/123/comment/comment-ok All-right! Got your comment. This is what we're going to save: "I love your post. Viagra cheap! -- Alex" Click the link below to save. => save 30 gemini://example.org/story/123 20 text/gemini\r Bla bla bla...
---