April 2005 - Development - lists.drupal.org

[drupal-devel] [task] Make status values in comment.module defined constants for readability
by robertDouglass 25 Apr '05

25 Apr '05

Issue status update for http://drupal.org/node/21221 Project: Drupal Version: cvs Component: comment.module Category: tasks Priority: normal Assigned to: Anonymous Reported by: robertDouglass Updated by: robertDouglass Status: patch Attachment: http://drupal.org/files/issues/comment_constants.patch.txt (9.89 KB) Reading the comment.module is more confusing than necessary because of all the comparisons $status = 0 or $status = 1. What does this mean? If one were to go by the example in node or users, 0 would be unpublished. Not so with comments, thus the confusion. The fix is to define constants and use them instead of 0 and 1. That's what this patch does. robertDouglass

4 6

[drupal-devel] [feature] Enable multiple block regions (not just "left" and "right" sidebars)
by chx 25 Apr '05

25 Apr '05

Issue status update for http://drupal.org/node/16216 Project: Drupal Version: cvs Component: block.module Category: feature requests Priority: normal Assigned to: Anonymous Reported by: paragkenia Updated by: chx Status: patch Let's get back to the code. If my understanding is correct the big discussion is about how to name potiental rectangular areas _after_ this is commited. So, first get this committed. What's need to make that happen? chx Previous comments: ------------------------------------------------------------------------ January 26, 2005 - 11:27 : paragkenia I read the comparision discussion between *Drupal* and *Mambo*. In several messages it was outlined that Drupal can place blocks only in right and left and not flexible to put them on anywhere where one want. It will be great if this can be changed in upcoming versions. I am no pro at PHP, so don't know how much time this task will take, but I think it is very important. parag ------------------------------------------------------------------------ April 15, 2005 - 02:44 : nedjo This issue was apparently partially addressed in issue http://drupal.org/node/19694 [1]. [1] http://drupal.org/node/19694 ------------------------------------------------------------------------ April 17, 2005 - 01:24 : nedjo Attachment: http://drupal.org/files/issues/block-dynamic-regions.patch (8.17 KB) This much-requested functionality - to have the ability to place blocks in more than the two predefined regions - was partially addressed in issue http://drupal.org/node/19694. [2]. But "blocks" are still limited to the "left" and "right" sidebars (hard-coded in block.module). This patch is a first step designed to enable multiple (eventually, admin-definable) regions for blocks. I've moved the existing "left" and "right" block regions to a 'region' table (with ids of 0 and 1, as currently used in themes). Then all references to the regions are drawn dynamically from the table. This way, if further records are added, they will appear in the list of available regions for block placement. Doing this actually reduces some duplicated code, since it's no longer necessary to repeat code blocks for each of "left" and "right". As it stands, the patch doesn't add any new functionality--but I don't think it breaks anything either. New functionality would need (a) new regions defined, and (b) changes to themes. A simple first step might be, e.g., to add a "footer" region and then add a call in the footer generation to append any blocks assigned to the footer region there. I'm setting this to patch, but I'm aware that it needs some discussion and refining before it'll be ready to apply. [2] http://drupal.org/node/19694. ------------------------------------------------------------------------ April 17, 2005 - 02:05 : nedjo Attachment: http://drupal.org/files/issues/block_regions.png (5.65 KB) Here's a screenshot showing the block admin page, with drop-downs for region placement (the options are dynamically generated based on defined regions). ------------------------------------------------------------------------ April 17, 2005 - 03:57 : adrian The biggest problem with this is that you can have multiple themes, and each of these themes can have different regions available. Also, the method of defining which regions are available needs to be standardised. Some of the work that me and Vlado are working on for the install system would go towards solving that problem (ie: meta information for modules, themes and styles). This has been discussed to death, but the general consensus has been that we _want_ to do this, but we need to solve a few other problems properly first, the most pertinent being the interface issue. Chris (factoryjoe) recently did a whole mess of workflows for something related to this. ------------------------------------------------------------------------ April 17, 2005 - 05:09 : syscrusher The first paragraph of adrian [3]'s post is a point that occurred to me also, as I read this thread. One suggestion would be to separate the definition and configuration of blocks, on one hand, from the placement of those blocks, on the other hand. In other words, Drupal core provides a mechanism defining what blocks exist, which of these are on by default or off by default and user-selectable vs. which are forced on for all users, and the configuration (if applicable) of specialized blocks defined by particular modules. Each theme provides a standard hook function that returns an array of region names and help/description text, e.g., array('left'=>t('This vertical region is left of the main content area'), 'right'=>t('This vertical region is right of the main content area'), 'footer'=>t('The footer is below the left, right, and main content areas of the page')). The theme part of Drupal core (i.e., theme.module itself, not the individual themes) provides a standard UI that is displayed within config of *each theme* (but is one physical code base within theme.module) that allows the administrator to map blocks defined by Drupal core into regions defined by the theme, and storing that mapping as an theme-to-block_ID-to-region_ID (with weight) table in the database. From there, the actual page rendering is similar to what's being done now, but there is more of it. I guess what I'm trying to say is that the key to solving this problem is breaking it along its degrees of orthogonality, and there are three -- two in Drupal core and one in the individual theme. Scott [3] http://drupal.org//user/1517 ------------------------------------------------------------------------ April 17, 2005 - 08:14 : Dries Scott, is right. The theme should export its available regions. Then the administrator's task is to assign blocks to regions (not to setup regions). A theme could have configurable regions, but that internal to the theme. To me, the real challenge is the UI and the interaction design. Configuring blocks could easily become a real pain (while most themes continue to use 'left' and 'right'.) Of course, we can implement all the functionality, default everything to 'left' and 'right', and worry about the UI later. This should be fairly straightforward to implement and nedjo's patch looks like a first step in the right direction. ------------------------------------------------------------------------ April 17, 2005 - 22:09 : nedjo I agree with the suggested approach of themes registering their regions. I'd see this happening when a particular theme is activiated (or upgraded). Regions would be an array of names (e.g., 'left', 'right', 'footer'). New records would be created in the region table only for region names not previously registered. Some areas I'm not clear on, or that need further work: Table naming Should a new table be 'region' or 'regions' (I've used 'region')? I don't see a convention among existing table names, some of which are singular and others plural. Default values I've kept the existing '0' and '1' ids for regions (left and right, respectively), for backward compatibility. But this means we can't use autoincrement for the region id (rid), since autoincrements seem to begin with 1. Likely we should change to autogenerated ids. Initial records Using the INSERT INTO statements as I've done for the initial region entries is probably counterproductive, since sooner or later they'll have to be dynamically generated. I was hoping this could be a preliminary patch, with the main work coming later, but likely we need to solve region registration by themes before this patch is applied. I don't have a good idea of how region registration by themes would be implemented (a hook on theme activation?), and invite suggestions or implementations. Theme system changes Besides region registration, the other change I'm seeing that would be needed in the theme system is loading blocks by region name, rather than region id. This is because the ids currently used ('0' and '1') in theory might be diffferent on a particular install. ------------------------------------------------------------------------ April 17, 2005 - 22:18 : syscrusher Instead of having themes "register" their regions, why not just add a theme API hook called them_regions() that returns an associative array of $region_name=>t($region_helptext)? This would be in keeping with other similar API functions, such as those that return what permissions apply to a module or what node types are defined by a module. Most themes are going to define only a small number of regions (two being the typical case now, but I could see five or six in a complex theme), so returning a constant array will be faster than querying SQL to obtain an array of rows. There could be (optionally, at the discretion of whoever builds this thing) another API hook like theme_region_properties($region_name) that returns an associative array with detailed info for a given region, such as extended help text with recommended usage instructions for the region. Scott ------------------------------------------------------------------------ April 17, 2005 - 22:19 : syscrusher s/them_regions/theme_regions/g ------------------------------------------------------------------------ April 17, 2005 - 23:47 : adrian Because. The most common 'theme' is a phptemplate theme. And there needs to be a generic method for specifying the regions available, in a non hook_function format.. Themes get their names from the directory in which they are contained.. when copying the theme to another directory, there must be _no_ modifications necessary to allow normal usage. This is one of the tenets of the new template system I designed. You would need a standard way to define meta-information for themes, that does not need to be modified when copied to a new directory. We are working on this in the install system work, as you need a way external of Drupal to define the module dependencies and some other things. My approach would be to add a theme.dsc text file to the directory, which would allow you to specify meta-information. For instance : ---- regions: left, right, footer, center, mountie author: Johnny McAngstyPants ---- ------------------------------------------------------------------------ April 18, 2005 - 03:00 : syscrusher Adrian wrote: " Themes get their names from the directory in which they are contained.. when copying the theme to another directory, there must be _no_ modifications necessary to allow normal usage. This is one of the tenets of the new template system I designed. " A very good point. But region names don't have to be unique across different themes, so my hook function wouldn't have to be modified. The suggestion I made for the mapping metadata was three factors: theme ID or name, region ID or name, and block ID (plus one non-key weight value to order the blocks within a region, but this is not relevant here). I'm not familiar enough with PHP template to be able to respond to your comments on that one, so if you say it's not feasible to work with my schema, then I'll take your word for it. :-) Scott ------------------------------------------------------------------------ April 18, 2005 - 03:03 : syscrusher Oh, wait.....I see now what you mean! It's not the output of the function that is the problem, but the *name* of the function. Never mind. I concede your point. Scott ------------------------------------------------------------------------ April 18, 2005 - 08:44 : Jaza I'm probably jumping ahead a bit here... but something that this patch will have to eventually account for, is how to allow regions to be defined as inline. Allowing a theme to define a region's position as being anywhere on the edge of the page (i.e. top, left, right, bottom, corners, etc) is relatively easy. But what about having a region that's in the middle of a node's body, or somewhere else that's not the edge of the page? What I'm thinking of, is eventually allowing the custom region system to integrate with the CCK. Like I said, I admit that I'm jumping ahead into the future - the CCK still has a long way to go before it's ready. But ultimately, it would be cool if a theme could define a region as being after, say, the "rating" field in nodes of type "movie review". This would make the region (and the blocks in it) truly inline and in context with the actual content of the node. This would be a huge step forward compared to the current block system, which doesn't allow you to mix the presentation of blocks and nodes at all. In Drupal ATM, blocks and "the rest of the page" are totally separate, and really you have no choice but to display them as such. The result of this is that information in blocks that really should be displayed as part of a node, gets literally "pushed to the side", and always seems secondary to the actual content. Often the block has content that is just as important as the content of the node itself. Another option (of less merit, IMO - because of the extra maintenance work, and lack of enforced consistency) would be to have a region filter. You could then choose where to place an inline region on a per-node basis, by entering text such as [region:2] (example based on image_filter syntax) into the node body. Anyway, just thought I'd throw that idea into the open. I don't expect it will be implemented any time soon, but it's something to think about. Jeremy ------------------------------------------------------------------------ April 18, 2005 - 09:54 : stefan nagtegaal Attachment: http://drupal.org/files/issues/regions---possibility-1.png (1.29 KB) Imo the 'regions' are more than you guys name here.. FOr example I attached 2 screens which only shows you the regions.. What I think is that it should always be possible to have free names for the regions. No matter if i call - the region where my content appears in - 'content' or 'site items'.. See attached screens.. ------------------------------------------------------------------------ April 18, 2005 - 09:56 : stefan nagtegaal Attachment: http://drupal.org/files/issues/regions---possibility-2.png (1.39 KB) And the second... See these links: - http://drupal.org/files/issues/regions---possibility-1.png - http://drupal.org/files/issues/regions---possibility-2.png ------------------------------------------------------------------------ April 18, 2005 - 16:22 : nedjo Attachment: http://drupal.org/files/issues/block-dynamic-regions2.patch (13.39 KB) Here's a revised patch implementing some of the ideas so far. I've used the theme engine - rather than individual themes - to generate the list of available regions. I know this isn't ideal, but e.g. for xtemplate it's at present the only option--no logic is possible in the individual theme. region is changed to a varchar field from the current tinyint. I've roughed in several regions--really just for demo purposes. More thought and work would be needed to refine them (e.g., styling, etc.). But this maybe moves the conversation forward. ------------------------------------------------------------------------ April 18, 2005 - 22:26 : Dries This patch enables different regions to be used. That is a good thing. The next question is: should we take into account that different themes could assign blocks to different regions? Lots of problems would be solved if (i) there could only be one active theme on a website or (ii) if all themes would have a common set of regions. (This "let users select their own theme"-thing is a left-over from the early days, except for WAP stuff maybe.) ------------------------------------------------------------------------ April 18, 2005 - 22:37 : killes(a)www.drop.org Giving up multiple theme support would be a great thing. People could still select from multiple style sheets. WAP shoud not be a user preference setting, but automatically detected and redirected to a special site with a WAP theme. ------------------------------------------------------------------------ April 18, 2005 - 22:48 : adrian You would then lose the ability to make a special front page theme for use with the sections module, for instance. Also, currently styles exist in the same namespace as themes, and we would have to make more a distinction of styles vs templates (or themes), but I do think this could help simplify things. ------------------------------------------------------------------------ April 21, 2005 - 17:54 : nedjo Attachment: http://drupal.org/files/issues/block-dynamic-regions3.patch (15.69 KB) The main issue raised, if I'm understanding well: if a theme or theme engine is present that offers regions other than those available in the default theme, any blocks assigned to those regions will be invisible to most users. Here's a revised patch addressing the concern. I've added a test for each region to see whether it's also available in the current default theme. (Actually, I'm realizing as I write this, I've used the theme engine--so it would need to be updated to test first if an engine is being used.) In any case, any regions not available by default are starred, and a message text appears (only if necessary). If there's support for the approach, I'd be happy to finish the patch. Remaining work: * extend to work with non-engine themes * finalize base set of regions to implement * implement common region set in all core themes (for now I've only done xtemplate) * work on CSS display (how blocks in each region should look) This last point I'm not too confident on and would look for help/suggestions. Screenshot to come. ------------------------------------------------------------------------ April 21, 2005 - 17:56 : nedjo Attachment: http://drupal.org/files/issues/block_regions_message.png (12.34 KB) Screenshot showing starring of non-default regions plus message. ------------------------------------------------------------------------ April 21, 2005 - 18:19 : killes(a)www.drop.org I like that approach. +1 ------------------------------------------------------------------------ April 21, 2005 - 18:23 : resmini My two cents. If this has to be done (and it should), better get rid of spatial definitions for areas such as left, right, top, bottom, etc. Use semantic names related to content and let the theme take care of their positioning interely. Also note that there in an ongoing debate in IA trying to build up a standard naming convention for page regions in a way that is semantically (and logically) correct. Just for the sake of it, you can check http://www.stuffandnonsense.co.uk/archives/whats_in_a_name.html http://www.stuffandnonsense.co.uk/archives/whats_in_a_name_pt2.html with follow-ups, including Eric Meyer's. There are more substantial contributes in the AIfIA web site, but I can't seem to find them now. I'll post a link or an abstract if I manage to. -- vector at exea dot it ------------------------------------------------------------------------ April 21, 2005 - 19:15 : nedjo Good points, useful references. The default regions I'm thinking of are: 'banner' => t('banner'), 'left' => t('left sidebar'), 'right' => t('right sidebar'), 'body_top' => t('body top'), 'body_bottom' => t('body bottom'), 'footer' => t('footer') Perhaps "body" would be better as "content"? It would be easy enough to add in more regions, e.g., within the content (after title, after help, etc.) One issue is that many of these regions will already have rendered content. Should the blocks come before or after that other content? I've assumed after for banner and footer, while giving both options in body. In any case, refinements or suggestions would be great. ------------------------------------------------------------------------ April 21, 2005 - 19:56 : resmini Well, 'banner' => t('banner'), 'left' => t('left sidebar'), 'right' => t('right sidebar'), 'body_top' => t('body top'), 'body_bottom' => t('body bottom'), 'footer' => t('footer') The classic (as far as classic goes for something Web related) scheme comprises 5 spatial regions, much like stefan's http://drupal.org/files/issues/regions---possibility-1.png with a left area mirroring the one called 'blocks'. Call them top, left, center, right, bottom. Variants usually exclude one or more of these, layout-wise. If we talk content (semantic) instead, things get a bit more complicated, as these spatial regions normally include more than one content-area. The top, for example, could include ad banners and the site's header, or the main menu. But these merge seamlessly with CSS definitions, which are definitely something that should get properly standardized (most of the errors / inconsistencies I'm encountering with Drupal today are related to this), but this is out of scope in this thread. What I somewhat object to the definitions above is that it mixes the two approaches. 'banner' is not spatial, is content, as it is the distinction between body top and bottom. If 'banner' goes there, it is 'top', whatever that might be. I designed a couple of web sites which had headers/logos/banners at the bottom of the page, and I'm not exactly the wildest designer you'll probably meet. And if we include body_top and _bottom, we should also add 'navigation', or 'menu' and some other content-related semantic region (impressum, or company_data, or whatever). I suggest that either we go along 'top' => t('banner'), 'left' => t('left sidebar'), 'center' => t('body'), 'right' => t('right sidebar'), 'bottom' => t('footer') or we dedicate some time to detailing what could possibly fit there semantically, which is perfectly possible without limiting design freedom or whatever, since the vocabulary is not infinite, but not something you do out of the blue. Please note that I understand t('something') to be an example, since such a layout only defines regions on the page, not semantic regions. My main content could fit nicely in the bottom region. As for the names for the regions, they are not important, as much as they are coherent and enforced. ------------------------------------------------------------------------ April 21, 2005 - 22:16 : stefan nagtegaal Attachment: http://drupal.org/files/issues/regions-final.png (2.39 KB) Maybe something like this??? But, I have to say that I _really_ liked the solution to split the 'content'-region into the 'content' and 'beneath content'.. I love that! ------------------------------------------------------------------------ April 21, 2005 - 23:20 : resmini Stefan, yes. That's basically what you can build if you stick with spatial, and this layout can generate quite an amount of (sane) layout variations even if you interpret it strictly (suppose a default) as <div id="header"> </div> <div id="left-bar"> </div> <div id="content"> </div> <div id="right-bar"> </div> <div id="footer"> </div> What you can do with this is left basically to the theme designers and their use of CSS, CSS-P or tables (brrr), but this could be an easy and proper way to identify macro-regions in which more accurate positioning happens at a more specific level. I'm all in favour of subsequent identification of sub-areas (top and bottom in body), but I don't think it would be a good idea to mix the different approaches I mentioned in my previous post. A site with lots of things going on or which is not just blogging surely has the need to use its content area in a wiser manner than just putting there 'content', the number of modules doing precisely this is more than proof, but that falls more in a 'Drupal definitive guide to semantics', in which CSS selectors are used or suggested strategically for such a scope. >From this point of view and for the sake of this thread, any unique area / div (#name) could (should?) be a target for blocks, and this in my opinion (*) would be the perfect solution, but I do not know if and how this is even remotely possible in the current Drupal. (*) It's quite late, local time, and it was a looong working day. ------------------------------------------------------------------------ April 21, 2005 - 23:38 : resmini Attachment: http://drupal.org/files/issues/regions-final-variation.png (1.9 KB) Just to be clear, I add a variation of stefan's scheme that shows how this is *not* by any means positional, but spatial, or better relational. Use your CSS imagination to take out regions, reduce them to empty containers, enlarge them, make them taller, shorter, whatever. 5 regions, do what you like. ------------------------------------------------------------------------ April 21, 2005 - 23:44 : resmini Sorry, I forgot. And this is also why I'm not too keen on calling anything 'content'. If I enlarge my right region enough, why not put my content there? Shouldn't this be a theme-level concern? It's nothing serious, but if my center area is called 'center' (or something similar and more sexy), I won't confuse anybody. This is left, center, right: do with them what you prefer. Content in the bottom? Show me. If I call something 'content', I'm more than suggesting that content goes there, and only there. ------------------------------------------------------------------------ April 22, 2005 - 09:17 : stefan nagtegaal Resmini, maybe it is only me but I think that you are making a mistake in your way of thinking.. I get the feeling that you're mixing the 'regions' and 'divs'.. the 'regioins' should have meaningfull sentences, but the divs shouldn'thave.. As an example: $region_name => $content_body here we go: 'header' => $logo 'left sidebar' => theme('blocks', 'left') 'right sidebar' => theme('blocks', 'right') 'content' => $content 'footer' => $footer. So, $region_name does not have to be equal to the id or class of the divs.. ------------------------------------------------------------------------ April 22, 2005 - 10:27 : resmini Stefan, >So, $region_name does not have to be equal to the id or class of the divs.. Yes, absolutely. I probably didn't make myself very clear: what I'm trying to point out is that in 'header' => $logo 'left sidebar' => theme('blocks', 'left') 'right sidebar' => theme('blocks', 'right') 'content' => $content 'footer' => $footer. there is no reason at all to call 'content' the region that holds $content. nedjo asked in #24 'body_top' => t('body top'), 'body_bottom' => t('body bottom'), if /Perhaps "body" would be better as "content"?/ and all my reasoning is simply that no, that is not content, or it shouldn't be (or better: it may happen that main content is elsewhere). As a matter of fact, even the connotation 'sidebar' doesn't fit very well logically. To make it short: Today's method works, but has limits. Mambo, which paragkenia quotes at the beginning, handles more 'user regions' but in pure Mambo style conventions are so loose that they are hardly useful. If we define regions, as in nedjo's example, in my opinion these regions shouldn't imply informational layouting, only relational positioning (that is, where they stay in the browser viewport). I know it may look like overdesigning, but Drupal constantly enforces conventions and coding guidelines everywhere: this is one of the best aspects of the project and I think it should carry to the layout. ------------------------------------------------------------------------ April 22, 2005 - 10:32 : stefan nagtegaal While I think that 'body_*' is not as self explaining as 'content_*' I vote for using 'content_*'..

1 0

[drupal-devel] [bug] Node options are still looking for individual variable instead of an array
by Prometheus6 25 Apr '05

25 Apr '05

Issue status update for http://drupal.org/node/20868 Project: Drupal Version: 4.6.0 Component: blogapi.module Category: bug reports Priority: normal Assigned to: Anonymous Reported by: Prometheus6 Updated by: Prometheus6 Status: patch Attachment: http://drupal.org/files/issues/BLOGAPI_4_6.txt (1.17 KB) blogapi.module can create (or attempt to..) any node type in 4.6, but the code to create a new node still assumes the node will be of type 'blog'. In addition, it's still looking for individual variables rather than the arrays that are stored in 4.6. It initializes those fields with calls like variable_get('node_promote_blog', 1); This will always return 1 unless you've upgraded to 4.6 from a previous version and set your blog nodes not to publish by default, and it will use that value for every type of node. This patch retrieves the array (and "comment_$node->type") variable for the appropriate node type and assigns the values accordingly. Prometheus6

3 3

[drupal-devel] [bug] aggregator broken by the theme('page') patch
by walkah 25 Apr '05

25 Apr '05

Issue status update for http://drupal.org/node/21282 Project: Drupal Version: cvs Component: aggregator.module Category: bug reports Priority: critical Assigned to: walkah Reported by: walkah Updated by: walkah Status: patch Attachment: http://drupal.org/files/issues/aggregator-print-theme.patch (2.09 KB) the recent print theme('page') patch broke aggregator pages (which had their print theme('page') calls nested). patch attached. walkah

1 0

[drupal-devel] no 4.6 for Drupal Docs?
by Anisa 25 Apr '05

25 Apr '05

Feel free to yell at me for not searching, but this seemed like an odd omission. Head has already opened, right? Anisa.

1 0

[drupal-devel] [bug] prefix.sh need fixes with postgresql
by patryk 25 Apr '05

25 Apr '05

Issue status update for http://drupal.org/node/21246 Project: Drupal Version: 4.6.0 Component: other Category: bug reports Priority: normal Assigned to: patryk Reported by: patryk Updated by: patryk Status: patch Attachment: http://drupal.org/files/issues/drupal-cvs-prefix-pgsql-fix.diff (557 bytes) prefix.sh does not work correctly with PostgreSQL database scheme. patryk

2 2

Re: [drupal-devel] DrupalForge?
by puregin 25 Apr '05

25 Apr '05

On 25 Apr 2005, at 12:53 AM, Djun Kim wrote: > > On 25 Apr 2005, at 12:29 AM, Jeremy Epstein wrote: >> > >> mamboserver.com -> apple.com >> drupal.org -> lego.com >> >> Take a look at the pairs of sites side-by-side, and you'll see what I >> mean. The similarities are mainly in the choice of colour. Maybe the >> Drupal community should look into making little plastic toys, just as >> a fundraiser on the side? :-). Clean, modular, accessible, semantic, >> usable, and [free... err, I mean...] reasonably priced plastic toys, >> of course. > > Lego is almost as famous for its lawyering as for its plastic. I > think > we should be careful ;) > > -- > puregin >

1 0

[drupal-devel] [feature] Improve search link ranking
by Steven 25 Apr '05

25 Apr '05

Issue status update for http://drupal.org/node/21152 Project: Drupal Version: cvs Component: search.module Category: feature requests Priority: normal Assigned to: Steven Reported by: Steven Updated by: Steven Status: patch Attachment: http://drupal.org/files/issues/search.link.ranking.patch (2.87 KB) Existing system Search.module currently tracks links to nodes, and indexes the words in the link's text as part of the target node. Thus, if someone links to the magic quotes FAQ with <a>more info about magic quotes</a> then the target node will get a higher ranking for a query containing any of the words "more info about magic quotes". However, there are two problems with this: 1) Most people don't provide a good caption for their link ("click here" or even just the URL). 2) Sometimes there are a bunch of names for a particular subject and the user used a less common name (e.g. teaser, excerpt, summary). This means that the link hint of "this page is important" often does not come into play simply because the search query doesn't match any linked keywords. The search query goes like this: SELECT i.sid, i.type, SUM(i.score / t.total) AS score FROM search_index i INNER JOIN search_total t ON i.word = t.word WHERE (...keyword matching...) GROUP BY i.sid, i.type ORDER BY score DESC Thus it selects all matching keywords for the query, groups them by source (a sid/type pair, e.g. "node" + nid) and sums the (relative) scores per source. For example when searching for "Drupal 4.6.0" one might get: #wordsidfromsidtypescoretotalsum (per sid/type)#1drupal1NULLnode6106/10 + 2/5 + 1/10 = 1.1#24601NULLnode25#3drupal13node110#4drupal2NULLnode3102/10 = 0.2 #3 is a link from node 3 to node 1 with the word "drupal". The other rows represent words that appeared literally in the result node. Note that these are the rows before grouping and summation. MySQL will only return two rows, where the first row combines #1,#2,#3 (node 1 with score 1.1) and the second row is only for #4 (node 2 with score 0.2). Changes This patch improves the results by assigning a de-facto score boost to nodes which are linked to. Only the node itself still has to match one of the given keywords, not the link texts. The node receives a boost proportional to how many times it has been linked to. It fits well within the existing search.module idea of trying to ensure the top results are relevant rather than focusing on making all results relevant. It does not obsolete the old mechanism, but complements it. Implementation-wise, this is a bit tricky. Issues The new, generic link boost should however only be applied once per item, not once per keyword. And if an item does not match any keywords, we don't want it to show up in the results (even if it has links pointing to it). There are two approaches to this: - Store the link boost among the regular words with an empty word as a special marker. When selecting the results, we assume the empty word is implicitly part of the search query, but we need to only keep those sid/type pairs that have at least one non-empty word in them. The only way I can think of is to require that each indexed item has a matching row (with a 0 score if there are no links). Then we can do a COUNT(*) per item and require that each sid/type pair has at least two matching rows (one link boost + one real word). This would of course be quite wasteful as there are nowhere near as much links as there are nodes. SELECT i.sid, i.type, COUNT(*) AS count, SUM(i.score / t.total) AS score FROM search_index i INNER JOIN search_total t ON i.word = t.word WHERE (i.word = '') OR (...keyword matching...) GROUP BY i.sid, i.type ORDER BY score DESC HAVING count > 1 - Store the link boost in a separate table to join on. As far as I know it's not possible to do one join per GROUP BY item, so I'd have to use a MAX() statement to reduce the link boost which is fetched per keyword to one score per sid/type pair. If we don't want an entry in search_links per item, then we have to use an if() statement to see if the join succeeded (0.02 is the score boost per link): SELECT i.type, i.sid, (SUM(i.score/t.count) + MAX(if(j.sid IS NULL, 0, j.score))*0.02) AS score FROM search_index i INNER JOIN search_total t ON i.word = t.word LEFT JOIN search_links j ON i.sid = j.sid AND i.type = j.type WHERE (...keyword matching..) GROUP BY i.type, i.sid ORDER BY score DESC In this patch I implemented a hybrid option: the link boosts are stored in the search_index table with an empty word, like in option 1. But like option 2, we use a join (on search_index itself) to fetch the boost and we don't store link counts for items without links. SELECT i.type, i.sid, (SUM(i.score/t.count) + MAX(if(j.word IS NULL, 0, j.score))*0.02) AS score FROM search_index i INNER JOIN search_total t ON i.word = t.word LEFT JOIN search_index j ON i.sid = j.sid AND j.word = '' WHERE (...keyword matching...) GROUP BY i.type, i.sid ORDER BY score DESC The reason I used the same table is because the links table would otherwise simply have the same DB structure as the search_index table. The word column is keyed, so restricting to j.word = "" should be fast. While this patch works, I'm not too sure about the implementation. I'm not an SQL guru, so feedback is certainly needed on all aspects. Results I can't really try it out on Drupal.org because it requires re-indexing, but I do have some clues that it would provide an improvement: of all the linked keywords in the search index now, 1977 are for the word "http", which almost certainly means a literal URL. Thus these records are almost never used, unless someone actually searched for "http", even though they tell us something useful (the linked nodes were interesting). Steven

2 1

[drupal-devel] [task] Consistent return 'passive voice' in form_set_error() and drupal_set_message()
by puregin 25 Apr '05

25 Apr '05

Issue status update for http://drupal.org/node/15595 Project: Drupal Version: cvs Component: base system Category: tasks Priority: normal Assigned to: stefan nagtegaal Reported by: bertboerland(a)www.drop.org Updated by: puregin Status: patch Attachment: http://drupal.org/files/issues/drupal_set_message_new.patch (50.98 KB) The attached patch updates Stefan's earlier patch with a few corrections and tweaks to messages. The bulk of these are minor typos, corrections to noun/verb number agreement, consistent use of 'could not' vs. 'can not', etc. In includes/locale.inc, I have made the tone of messages slightly more formal by replacing '... file is broken, because...' in messages with '... file contains an error: ...'. While visiting the affected files, I also made minor fixes or improvements to other messages. I have re-written portions of the help text for locale.module. Please review and comment! :) puregin Previous comments: ------------------------------------------------------------------------ January 14, 2005 - 05:59 : bertboerland(a)www.drop.org *My* drupal not *Yours*! I think that when one addresses the user or the administrator of a site, it is best to do this outside in instead of inside out. So when in the documentation the coder is referring to the site of the administrator, this site should be called "your site" (/inside out/) but "my site" (/outside in/). There are numours examples where the text in referring to "your"where "my"would be apopriate [1] There used to be some "your" in the userinterface for the enduser as well, but that has been solved in 4.5. Before that you had "my account", but "track your postings". This has now all beens replaced by "my"and "mine". If everybody agrees, lets also corect this for the administrator as well; my administrator, not yours ;-) [1] examples: admin/logs/referrers "...that point to your web site." admin/settings "General configuration options for your site" ------------------------------------------------------------------------ April 19, 2005 - 08:36 : stefan nagtegaal We should consistently use the 'passive voice', when returning messages using form_set_error() and drupal_set_message(). ------------------------------------------------------------------------ April 22, 2005 - 04:05 : stefan nagtegaal Attachment: http://drupal.org/files/issues/drupal_set_message consistency part 1 of 2 (51.16 KB) first patch of 2 to make the drupal_set_message(), form_set_error() and watchdog() messages more consistent. please review/comment and apply.. ------------------------------------------------------------------------ April 22, 2005 - 04:05 : stefan nagtegaal Setting status to Patch. ------------------------------------------------------------------------ April 23, 2005 - 00:49 : puregin A big -1 from me on replacing "you" with "me" :) Regarding the use of the first person - I find it extremely awkward and confusing to see usage such as 'My Account', "Here are my recent posts", etc. From my perspective as a native speaker of English, I find that this kind of usage violates conventional "conversational context". To clarify, if I am communicating with another person *via* the medium of a program then I expect that person to follow the normal conventions of conversation, in particular, to say "me, my, I" when referring to him/herself, and "you, your, etc." when referring to me. If I am communicating with a program, I *also* expect it to follow conversational conventions. Particularly obsequious programs may always refer to themselves in an indirect second person: "This unworthy program was unable to execute your most excellent command, and will henceforth terminate itself" I have to say I really don't like this kind of interaction. I like a program that tells me straight up "I can't find my configuration files! Sorry, I'm giving up!" People like, understand, and are extremely good at direct conversational interaction. It's intuitively clear who is speaking and what the relation to the listener is. I might tolerate a page with title "My Account", but really, a message such as 'This page shows my site-wide referrer statistics. I can optionally view just the "external referrers" or the "internal referrers". Referrers are web pages, both local and on other sites, that point to your my site.' is just plain confusing and unclear. To quote Sendmail: "I refuse to talk to myself" Please don't inflict this gratuitous use of first person on us! Djun ------------------------------------------------------------------------ April 23, 2005 - 05:28 : rivena Have to agree with Djun. You're going to run into a lot of problems changing all the you's to me's. It'd be creepy if my drupal install started talking about itself. Anisa. ------------------------------------------------------------------------ April 24, 2005 - 04:57 : Dries I checked a couple other sites like Amazon.com, Apple.com, NYT and they all use 'your'. I suggest we use 'your' and that we document this guideline. The exception is Yahoo!, who seems to use 'My Yahoo!'. ------------------------------------------------------------------------ April 24, 2005 - 10:30 : Bèr Kessels AFAIK its as follows: A site that offers services, will use " yours". Community sites will therefore use " your Profile". A site that is used by a single user, should use "my". Its My site, if I own that blog. But if I am but a part from a large community, they offer "your blog". Ber

1 0

[drupal-devel] [bug] Wrong GROUP BY statement crashes PostgreSQL
by patryk 25 Apr '05

25 Apr '05

Issue status update for http://drupal.org/node/21249 Project: Drupal Version: 4.6.0 Component: locale.module Category: bug reports Priority: critical Assigned to: patryk Reported by: patryk Updated by: patryk Status: patch Attachment: http://drupal.org/files/issues/drupal-cvs-locale_inc-pgsql-fix.diff (848 bytes) Wrong GROUP BY statement crashes PostgreSQL in locale.inc when exporting POT. patryk

1 0