….x.x.x-x

This optional means of debugging can provide a feedback loop when the
strict SQL mode is not enabled.

* collectMySQLWarnings(): new function to buffer the warnings
* usePreparedInsertBlade(): amend to call the above
* usePreparedDeleteBlade(): idem
* usePreparedUpdateBlade(): idem
* showMySQLWarnings(): new function to display the buffer
* index.php: call the above in the "redirect" case

Amend respective test to fail the right way.

* usePreparedDeleteBlade()
* usePreparedUpdateBlade()

This is how it was originally intended (also update the sample dataset).

* commitAddPortReal(): use nullIfEmptyStr()
* commitUpdatePortReal(): idem
* upgrade.php: queue an UPDATE for 0.21.0

* lastInsertID(): use PDOStatement::fetchColumn()
* getIPv4Stats(): idem
* getIPv6Stats(): idem
* getRackspaceStats(): idem
* sortPortList(): use array_fetch()

Convert a repeated code block to a function, fixup a TD alignment,
simplify portlet rendering, do not show the multiport form if
permissions would not allow to submit.

Address a long-standing glitch: when a port on the currently displayed
object was (or was suggested to be) linked to a port of an object that
had no name, the displayed remote object name would be an empty string.
This change makes the interface display a standard substitute instead,
which among other things makes the remote object hyperlink clickable.

* fetchPortList(): add remote object type ID to the result columns
* renderObjectPortRow(): make use of formatObjectDisplayedName()
* renderPortsForObject(): idem
* findSparePorts(): all of the above

For simpler maintenance the RackTables-specific CSS modifications now
reside in css/codemirror/rackcode.css. Do not include the CSS, HTML,
JavaScript, PHP and SQL CodeMirror tokenizers anymore as they had not
proved to be useful.

This completes the changes done in commit cf5da67.

* renderRackCodeEditor()
* renderTextEditor()

Reimplement the RackCode language mode in rackcode.js using simpler
JavaScript code and define a standalone "rackcode" theme in rackcode.css
instead of overriding the default theme. In renderRackCodeEditor() pass
more properties to CodeMirror constructor and make minor HTML fixups.
In renderRackCodeViewer() rewrite the code to use CodeMirror in read-only
mode and to scroll to requested line on request. Amend the URL format in
refRCLineno() respectively.

There is no reason to specify a default value for $input_name in
printTagsPickerInput() and printTagsPickerUl() so long as
printTagsPicker() is the only function that calls those functions.

A read-only RackTables instance would display an error image instead of
the thumbnail image for any rack that has invalidated thumbnail cache.
This change makes it deliver the thumbnail even if the attempted cache
update failed because of insufficient database permissions.

* RTDBTableAccessDenied: a new exception class
* convertPDOException(): add a respective case block for the error code
* dispatchMiniRackThumbRequest(): use the above for its REPLACE and
  update the comment

* getIPv4PrefixSearchResult()

Do the same for IPv6 and make a ChangeLog entry.

* getIPv4PrefixSearchResult()
* getIPv6PrefixSearchResult()

Keep it consistent with the other declarations in the file.

* updateObjectAttributes()

updateObjectAttributes() used to feed each new attribute value to MySQL
connection and let it raise an exception for an invalid input. The
exception would be dispatched as a hard error with the MySQL error
codes and a stack trace page.

Improve this by making calls to genericAssertion() as required and
converting any resulting IRAE exceptions to a human-readable form.

Add helper methods getValue() and getReason() to InvalidArgException.

In this context the class name is always PDOException.

formatAttributeValue() was initially purposed for objects and when it
yielded a hyperlink it would hard-code "page=depot" into it. When the
function was later reused for racks, rows and locations, this logic
didn't deliver working URLs.

Add necessary checks such that this feature only tries to work when it
can work, add a date format mouse hint, add minimal reporting of RackCode
parsing errors and move some code around for clarity.

Modify the functions below as required:
* renderRow()
* renderRackInfoPortlet()
* renderObject()
* renderSearchResults()
* renderLocationPage()

It used to be possible to map "date" attributes to racks, rows and
locations, but the properties editor would not know how to handle
those attributes in particular and would not make permission checks
correctly in general.

This change addresses those issues by replacing four similar code
blocks with one new function that can handle all types of attributes
for all types of objects and makes necessary permission checks.

* renderEditAttributeTRs(): a new universal function
* renderEditRowForm(): use the new function instead of custom local code
* renderEditObjectForm(): idem
* renderEditRackForm(): idem
* renderEditLocationForm(): idem
* updateObjectAttributes(): update request argument checks to match
  what the current implementation puts into the form

Specifically, not inside foreach.

* renderObject()
* renderIPForObject()

Original idea by Lucas Aimaretto.

Added link to SDK project page in error message when python package is not installed.

Use the same WHERE filter in subquery. Changed functions:
- scanIPv4Space
- scanIPv6Space

Remove an unused global, declare arrays before the first use, use foreach
instead of hard-coded index range, spell comparison with zero, eliminate
a redundant variable and use array_search().

Don't do the work permitted() already does internally and clarify control
flow. This eliminates two variables.

* renderRackspaceRowEditor()
* renderRackSortForm()

* proxyCactiRequest()
* proxyMuninRequest()

The input to sortPortList() is an array of arrays, if it is an array of
anything else the function (since commit 07d22ce) will use array_fetch()
on the elements anyway, causing issues downstream. Vladimir Sukhonosov
had reported numerous PHP warnings in this regard.

Fix usort_portlist() to feed proper argument into sortPortList().

If the argument is not an array, it is a bug in the code as arguments to
that function come from other functions, not directly from user input.
Raise an exception before the broken code makes any [more] damage. Add
more tests.

"If a file is pure PHP code, it is preferable to omit the PHP closing tag
at the end of the file. This prevents accidental whitespace or new lines
being added after the PHP closing tag, which may cause unwanted effects
because PHP will start output buffering when there is no intention from
the programmer to send any output at that point in the script."
-- PHP manual

There is no gain in assuming that if SNMP version is not 1, not 2 and
not 3, it means it is 1. Keep it simple.

There used to be a few code blocks that switched PHP interpreting on and
off with PHP tags, convert those to use either heredoc or plain echo as
that is the usual syntax for this purpose elsewhere in the code. Fixup
some indentation while at it.

* renderInstallerHTML()
* renderInterfaceHTML()
* renderIndex()
* dragon()
* renderSNMPPortFinder()
* renderPopupHTML()
* printStatic404()
* renderUpgraderHTML()

In fetchAttrsForObjects() throw an exception on an unknown attribute
type as that situation cannot have a reasonable default workaround.

is_numeric() can test numeric strings but it accepts floats as well
because that's the way it is defined to work; is_int() tests for an
integer. Combine those two to make isInteger() really do what it claims.

The contents of RackThumbnail.thumb_data used to be Base64-encoded but
the column itself does not require it as it is a BLOB so dismiss the
encoding and just store and read the image as it is. Delete all rows
from RackThumbnail.

* loadThumbCache(): rename to loadRackThumbCache()
* saveRackThumbCache(): new helper function for consistency
* dispatchMiniRackThumbRequest(): update to use the above

* isUnsignedInteger(): a new helper function that better matches the
  problem space than isInteger(), with tests
* assertUIntArg(): simplify by using the above
* setConfigVar(): the function would mistakenly accept float values for
  "uint" configuration variables, fix the check
* setUserConfigVar(): idem

This add colors to object depot, rack diagram, rack thumbnails and IPSpace.
The color depends on the tags assigned to an entity.
Each tag could be assigned one color. Multiple colors are displayed as a sharp gradient.

* listCells(), spotEntity(), getTagList(), commitUpdateTag(),
  renderTagRowForViewer(), printObjectDetailsForRenderRack(), renderRack(),
  renderDepot(), renderIPNetwork(), buildTagCheckboxRows(), updateTag(),
  printRackThumbImage(), renderTagTreeEditor(), enableTagsPicker(),
  getTagClassName(): add color
* objstyle_cache, color: new global variable
* isHTMLColor(), getColorSelect(), assertHTMLColorArg(),
  colorHex2Rgb(), getObjectClass(), setEntityColors(), destroyTag(),
  coloredObject(): new function
* genericAssertion(): add types "htmlcolor" and "htmlcolor0"
* rebuildTagChainForEntity(): clear RackThumbs on color change
* JS generateTagList(): add color class

MySQL char(6) can contain any string including a hex notation of a 24-bit
RGB colour. With this storage type the PHP code should check if the value
is valid every time it reads the column, and if the value is invalid
there is no good way to deal with it as the invalid value could have been
stored a long time ago and it could be caused by a bug in a
non-interactive script so there is nowhere to put the blame.

Based on the prior experience it is better to eliminate this place for an
error earliest possible, do this by switching to an unsigned 24-bit
integer type in the MySQL table, this way any non-NULL value stored in
the column is always valid and any attempt to store an invalid value will
fail at one or another level immediately so there is a feedback loop now.

* get_pseudo_file(): update CREATE TABLE
* upgrade.php: modify ALTER TABLE
* getTagList(): amend SELECT to convert to hex at server side
* commitUpdateTag(): amend column conversion
* $opspec_list: refine declarations for tagtree-edit-createTag
* isHTMLColor(): reduce to a small unary function and add tests
* HTMLColorForDatabase(): new converter function with tests
* HTMLColorFromDatabase(): idem
* assertHTMLColorArg(): simplify and merge into genericAssertion()

The previous commit passed the tests on PHP 7 but not on PHP 5 due to the
difference in is_numeric(). Make it work the same on both PHP versions
and add some more tests.

* rebuildTagChainForEntity()
* setEntityColors()
* buildTagCheckboxRows()
* coloredObject()

* getAllIPv4Allocations(): fetch one more column and fixup SQL formatting
* renderPopupIPv4Selector(): improve formatting like in commit f4d00fa
* sortTokenize(): don't call count() more often than required, use
  isUnsignedInteger() and numCompare()

All arguments to this function are mandatory, also amend the tests.

This way the code in doSwitchSNMPmining() and doPDUSNMPmining() is much
easier to understand.

This should fix the color SELECT in some browsers.

In the tag tree viewer apply tag class to the tag name only rather than
to the whole TR, this is consistent with the tag filter portlet and does
not conflict with the viewer sometimes using the TR to signal additional
information (a warning or not assignable) about the tag.

Firefox accepted "rgb(127.5,127.5,255)" but Chrome [properly] didn't:

"The format of an RGB value in the functional notation is 'rgb(' followed
by a comma-separated list of three numerical values (either three integer
values or three percentage values) followed by ')'." -- CSS spec

Add some tests.

...to give plugin devs more options to interact.

* string_insert_hrefs()
* email_regex_simple()
* getOptionTree()
* setEntityColors()
* renderRackSortForm()
* printRackThumbImage()

…Addresses

to allow plugin devs to extend/change the functionality of the IP list.

* pi.css
* renderDataIntegrityReport()
* renderObjectPortRow()

Switch some tables from hard-coded HTML to renderTableViewer().

With this change the script does not require the RackTables top
directory to be the current working directory anymore. Also it runs the
last query as root to avoid the password warning -- the next stages will
find it out anyway whether MySQL credentials work as expected.

move cacti and munin code into plugins

* getPlugins(): don't use "SELECT *"
* getPlugin(): idem
* formatPluginState(): use array_fetch()
* renderPluginConfig(): use renderTableViewer()
* renderPluginEditor(): idem and use "&&"
* installPlugin(): use the assertion result instead of a global
* uninstallPlugin(): idem
* upgradePlugin(): idem

Remove the PHP closing tag from the new PHP files.

Only show individual controls and the warning when respective permission
check would allow the action. This way what the user sees is much likely
to be what they can use.

* SQLDateFromDateStr(): new function with tests
* genericAssertion(): add respective case blocks
* resetUIConfig(): amend as necessary
* install.php: idem
* upgrade.php: idem

This is a new helper function, the current code base does not use it
right now but it may come handy later.

In the triggers section add the entries below to the list of known
triggers and compare the list with the items in the schema both ways.
Use renderTableViewer() to display the difference.

* EntityLink-before-insert
* EntityLink-before-update

In the foreign keys section make a similar change and amend the new
plugins to return their lists of FKs in the new format.

* RackThumbnail-FK-rack_id
* VLANDomain-FK-group_id

* renderSimpleTableWithOriginViewer(): use renderTableViewer()
* renderPortOIFViewer(): idem
* renderChapter(): idem and clean ref. counter processing up

RackThumbnail.thumb_data is not intended to be NULL, in the unlikely
case it is NULL just let the bug propagate and become visible elsewhere
so it can be fixed.

Like in commit f5b9c21.

Use in_array() instead of array_key_exists() to make the check actually
work. Do not return any result as it is wasted anyway. Eliminate a
temporary variable.

Move all read-only tests into one method to avoid duplicating the class
fixtures. Supply an array argument to usePreparedDeleteBlade() to make
use of SQL IN(). Assign some colours to the fixture tags and add a basic
read-only test to cover setEntityColors().

Make the state code in coloredObject() a mandatory parameter and update
printRackThumbImage() to make sure it is always properly set. This
restores the colour-based highlighting for objects that don't have any
colours associated through assigned tags. Reduce the footprint of the
colour list argument. Prune and rearrange some code for clarity.