The cockpit of the panel home made the site slow: every poll ran its own PowerShell probes, parallel polls ran them at once and each held the session of the administrator while it did. The server is now sampled by one writer under a lock into storage/logs/monitor.json, every poll lets the session go, and the load of a page becomes part of the statistic, so the panel shows its history by day and by hour. The layout of the site arrives with its own class.

Core changes:

1. One writer of the server sample (core/monitor.php, admin/modules/monitor.php, core/system.php):

2. getMonitorSample() retakes a sample older than ten seconds under the non-blocking lock monitor-sample.lock

   • a request that does not get the lock reads the file as it stands, so no probe ever runs twice at once
   • the cockpit poll and the HTMX polls of the monitor ask for a fresh sample, the monitor page and the

presentation only read

• the monitor job takes the same sample once a minute and fails when it is older than a minute

• the sample holds CPU, memory, disk, network and disk-io rates with timed histories, uptime, cores, the server software and the uploads size, retaken once an hour instead of on every monitor page

• monitor.json is written once per sample: the history functions take and return the sample
• the chart labels each point by its stored time; the live chip of the cockpit shows the age of the sample
• One PowerShell run per sample on Windows (core/monitor.php):

• getWindowsProbe() reads load, cores, clock, memory and boot time in one run through proc_open()

  • killed after ten seconds, so a hung WMI query never holds the lock
  • a run that gave nothing writes no sample: the old figures stay and age instead of turning into zeros
• every OS check is PHP_OS_FAMILY; stristr(PHP_OS, 'WIN') also matched Darwin, so macOS had no CPU figure
• the duplicate open_basedir closures of getCpuLoad() give way to isPathAllowed() and getFileSafe()
• the Linux CPU share is loadavg / cores 100; it was 10 and showed a full server at ten per cent
• the cpu_details and memory caches leave, the WMIC fallbacks send their errors to NUL
• Polls release the session (core/security.php, public/index.php):

• a request carrying X-SL-Poll calls session_write_close() right after the idle check

  • the cockpit, the queue bell, the private message poll and the HTMX polls hold no other request of the visitor
  • no poll handler writes the session after that point, checked handler by handler
• The load of a page in the statistic (core/system.php, admin/modules/statistic.php, statistic-extended.html):

• updateStatsTrack() adds generation (ms), queries and peak memory (KB) of the page to hour fields 17-19 of statistic.log, beside the hits by hour, so they roll into days.log and the month archives

• the statistic page shows the average page per day with a total row and an average generation by hour chart
• the strip of the cockpit shows the average page of the running hour instead of the figures of its own poll

• The layout of the site (core/classes/layout.php, config/layout.php, public/index.php, core/system.php, core/classes/template.php, core/classes/update.php):

• Layout reads config/layout.php: pages with their sections, the module slot, menus and the places of the theme
• the home page and the page of a module render through it; getBlocks() asks Layout::isViewShown() for the scale
• Template renders a section of the theme, sections/<type>.html
• an update adds new keys of layout.theme but never a page or a menu item the site deleted
• Help, documents and tests (admin/info, docs/ARCHITECTURE.md, docs/UPDATE.md, tests):
• the help of the monitor, the scheduler and the statistic describes the sample, the poll rule and the load columns
• CockpitTest holds the strip of the hour, LayoutTest the layout

Benefits:

• a cockpit poll costs 20-90 ms instead of 1.5-6.8 s, a page of the same session no longer waits behind it
• a sample costs about 1.7 s of PowerShell at most every ten seconds instead of several runs per poll
• the monitor page with expired extras renders in about 0.3 s instead of 4.3 s
• administrators see the load history of the site in the statistic

Technical notes:

• statistic.log gains fields 17, 18 and 19, twenty-four sums each; readers address fields by number, older days show a dash
• monitor.json gains cpu_p, cpu, mem, net, diskio, sys_hist_time, uploads and uploads_at
• the hours already counted on the day of the update carry no load and lower that day's average
• the rule that a poll handler never writes the session is documented, not held by a test

The stored texts are Markdown since the update converts them, so the writers stop escaping what they store and the parser stops reading the BB it no longer meets: every store keeps the typed text, the editor mounts it unchanged, the bracket layer reads [hide], [tabs], [attach], [usehtml], [usephp] and [block] alone. The full run of update.php on a copy of the production dump found two faults of the updater and the writers; both are fixed, the run is proven, and what lasts of it is reference in docs/.

Core changes:

1. One storage of a typed text (core/security.php, core/classes/, core/helpers.php, core/user.php, admin/modules/, modules/):

2. filterSourceText() turns a typed text into the stored source: CRLF to LF, the trusted tags, the censor list, never HTML-escaped, as the parser escapes at render

   • the blocks, the admin messages, the newsletters, the signatures and own blocks, the contact text, the forum, the

comments, the private messages and Node read the field raw and store through it

• Comment and Privat lose their own copies of the censor (filterCommentBody(), filterMessageText())

• getTplTextarea() mounts the stored source as it is: no <br> turned into a hard break, no entity decoded

• The forum, the signatures, the own blocks, the blocks, the admin messages, the newsletters, the contact text and the preview of the editor render safe with trust, as a Node material does

• The reply quote of a private message and of a forum post quotes the stored Markdown as it is

• The admin message form keeps an empty language; the snippet of a private message drops the quote markers of every depth and the escapes (filterSnippet() of Privat)

• The parser reads the SLAED tags alone (core/classes/parser.php, public/templates/*/fragments/, lang/):

• The handlers of the converted and removed BB tags, ed2k and the smilies go, with images/smilies/ of both themes and _SMILIE of the six lang files

• parser-inline.html, parser-block.html, parser-image.html and blockquote.html of both themes lose the parts only those tags drew

• The read-only editor (core/classes/editor.php, public/plugins/system/editor.js, fragments/editor-*.html, theme.css):

• Editor::getFrame() takes readonly; the frame marks its area, the kit marks every tool that changes a text with the need edit, and editor.js keeps CodeMirror read-only and without files

• .sl-markdown table of the admin theme draws the tables of the help
• The update (public/update.php):

• Points, ratings and fields whose mark is set and whose manifest is missing ran on the site before and answer done; they stopped the whole run on production, so no table was dropped

• The guard of getMigrateAttach() keeps an image inside a link to another site with text or a hard break beside it in the label, so a repeat no longer rewrites it

• The comment edit form of the panel (admin/modules/comments.php):
• The author and the IP are escaped once
• Help and documents (admin/info/, modules/contact/admin/info/, docs/):

• The help of the configuration, the blocks, the admins, the comments, the monitor and the contact text in the new state, the tables of the help readable

• docs/PARSER.md: the engine, the measurement, the grammar of the SLAED tags, what is no markup and the conversion of the BB era; docs/EDITORS.md: the stored text on mount and save

• docs/NODE.md: a site of an early 6.3 build, its folder moves, the address step of the migration, the columns no parser call renders; docs/UPDATE.md: moving slaed.net to 8.0, from the copy to the live check

• docs/VERSIONS.md: both changes
• Tests (tests/):

• ParserFixturesTest without the BB fixtures, EditorBreakTest on the storage of every writer, EditorFormatTest on the read-only frame, PrivatClassTest on the snippet

• UpdatePointsTest, UpdateRatingsTest, UpdateFieldsTest on a mark without a manifest, UpdateAttachTest on the label around a linked image

Benefits:

• A save without a change keeps every byte of a text; what an author types is what the store keeps
• One rendering path for every Markdown text of the site
• The move of slaed.net is a proven, repeatable run with its order written down

Technical notes:

• Breaking: [u], [color], [size], [family], the alignment tags, the alignment of an image, ed2k and the smilies - keep their text and lose the formatting; images/smilies/ and _SMILIE are gone

• No schema change; the stored texts were converted by the text unit of update.php

The places of a block listed the site modules alone, so a block could not be shown on the pages of one Node type: those pages carry the name of their type in $name, and the module node itself is never a page name. The list now offers every type beside the modules.

Core changes:

1. The places of a block (admin/modules/blocks.php):
2. getBlockWhereItems() lists the site modules and every type of config/node.php, sorted by name, and leaves "node" out
3. The popover of a box names what it is, module or type, with its code
4. Help (admin/info/blocks/ru.md):
5. The places list the modules and each type, and a block marked at a type shows on the pages of that type
6. Tests (tests/Unit/NodeIntegTest.php, tests/Support/route_probe.php):
7. The block form offers the type "docs" and no "node", and a save keeps the type the block was bound to

Benefits:

• A block meant for documentation or news pages shows there and nowhere else, without a module of its own

Technical notes:

• The stored value is the same list of names in _blocks.which; a type name is matched as the page name already is

The stored texts are moving off BB onto Markdown, so the parser that renders them has to read Markdown by the standard and not by its own habits. The examples of both specifications are now fixtures, and the parser was rebuilt on a block tree and an inline tree until every one of them passes in the trusted and in the safe mode.

Core changes:

1. Conformance fixtures (tests/Fixtures/spec/, tests/Unit/ParserFixturesTest.php):

2. commonmark.json holds the 652 examples of spec.json 0.31.2 as published, gfm.json the 24 extension examples of GFM 0.29-gfm (cmark-gfm 0.29.0.gfm.13), the two task list examples marked disabled there included

3. filterDocMeetsTheSpec() runs every example at $safe = false, and at $safe = true every one but those carrying raw HTML or a link of a scheme the safe rendering refuses

   • the comparison removes only the markup of the theme: heading ids, theme classes, the code figure, the

broken-image fallback, the cell alignment style and the disk check of a local image (Parser::$fileoff)

• the CommonMark examples run with the tag filter and the extended autolinks off (Parser::$gfmoff), the GFM ones

with them on, as the spec runner of cmark-gfm does; code is measured in the shipped highlighter mode 2

• failing.json is the baseline of examples that fail; a listed example that passes fails the test, so the list only shrinks, and it is empty: trusted 676 of 676, safe 591 of 591

• The engine (core/classes/parser.php):

• getBlockTree() reads the block structure one line at a time: indented and fenced code, HTML blocks of the seven kinds, link reference definitions, setext headings, lazy continuation, tight and loose lists, GFM tables

• getInlineTree() reads inline content with the delimiter and bracket stacks of the specification: flanking rules, code spans, references, autolinks, entities, hard breaks, GFM strikethrough and extended autolinks, mark

• getBlockHtml() and getInlineHtml() render the tree through the fragments; the regex passes filterCode(), filterBlocks(), filterList(), filterTable() and the Markdown tail of filterBbInline() are gone

• The bracket layer reads the copy of filterMask(), which a scan of the same parse masks; a raw pair whose opener the scan leaves visible is taken back whole, since its content is no Markdown

• A safe rendering keeps an address without a scheme and one of http, https or mailto; data:, javascript: and vbscript: die in every mode; a line of <br> opens an HTML block as any complete tag does

• Parentheses in a link target are bounded at 32 levels, as cmark bounds them
• Fragments of both themes (public/templates/*/fragments/):
• list takes has_start and start_num for <ol start>, parser-table drops <tbody> without body rows
• Documents (docs/PARSER.md):

• The standard, the comparison, the SLAED extensions on top of it and the bracket layer; the Markdown list names everything CommonMark and GFM define; the safe and the trusted mode as the engine reads them

Benefits:

• A text renders as every CommonMark and GFM tool renders it, which the conversion of the stored BB relies on
• The two known defects are gone: fenced code inside a quote or a list item, and <https://...> and <name@host>

Technical notes:

• Measured on the 28344 renderable texts of the stand: both parsers render them in the same time (8 to 10 s), the slowest text 43 ms, no warning; 6637 render differently, each by the standard

• The plain format keeps its own path; no render call of the site passes it

The production site runs a development build of early October whose Node texts point into the archive of the first migration and whose schema, settings and tables lag behind the release. update.php now brings such a site forward on a copy without migrating its content again, drops the tables of the modules gone for good and every source table whose rows are in Node, and carries jokes and media into Node for any other 6.2 site.

Core changes:

1. Catch-up of a 6.3 site (public/update.php, storage/update/sql/table_update6_3.sql):

2. setUpdateArchive() rewrites the uploads/archive/<module>/ and old folder addresses of Node texts and their comments into [attach] or the file address of the material, and an archive address elsewhere into the address of the published material that carries the name

   • a dry run (POST dry=1, "Count the file addresses") counts per column and writes nothing
   • a manifest with counts, cursors and state; a stopped run continues, a verified unit is skipped

3. setUpdateMerge() takes the retired keys out of a 6.3 site and adds every shipped option it lacks from storage/backup/release/config/; without that folder the step refuses

4. The eight time columns of the early-October tables reach the forms of table.sql through runiftype(), each moment kept; a 6.3 db.php is written in the collation of the release

5. Drop of the tables (public/update.php):

6. setUpdateTables(), at the end of the first stage, drops the tables of auto_links, clients, clients_down, money, order, partners, products and whois, and every source table of Node whose rows are all materials

   • a published row by its _node_legacy row, a pending row by a material of its type with its title and creation time
   • an empty table at once; a table with a row matched by neither stays, the report names up to ten ids

7. deleteMigrateTables() drops the table of a module right after its data step, once every row is a key of the map of the manifest and every material of the map is in _nodes; a manifest of another database keeps the table with a note

8. The page of the second stage shows a carried module from its manifest
9. Jokes and media (public/update.php, storage/update/sql/table_update6_3.sql):
10. Both join getMigrateMap(); "Batch X" renames their 6.2 columns

11. The twelve own columns of media land in the extra fields of its profile, every http(s) address of its links becomes a resource of the download role, ed2k and other schemes leave a note

12. The rating balances of both are carried
13. Text of the second stage (lang/*.php):
14. _NODE_MIGINFO says that the table of a module is dropped once its materials are carried, in all six locales
15. Tests (tests/):

16. UpdateArchiveTest with the archive unit, the settings merge and the drop on a site of early October (a cp1251 table, a pending row, a row no address names)

17. The update62 fixture gains the 6.2 tables of jokes and media with a rated joke and a film; UpdateSiteTest checks their materials, fields, resources and ratings, the tables left after each stage, and a stale table that stays

18. The probe adds its help request before the first run, because an empty source table drops there
19. SchemaUpdateValidationTest skips jokes and media like the other source tables
20. isTreeSkipped() no longer enters uploads/, whose copy of a site carries the PHP of old releases; TreeWalkTest holds it
21. Reference (docs/NODE.md):

22. The run order, the schema file, the tests and the migration of the removed modules describe the catch-up, the drop and the two new modules

Benefits:

• The production database moves to the current code keeping every row and id it holds
• No table of a gone or carried module is left behind, and none is dropped while a row of it is not in Node

Technical notes:

• Drops tables with no archive: the gone modules for good, the source tables only once their rows are materials
• update.php remains the maintainer's tool and is not shipped in a release