This file documents all changes in the profiler gecko and processed formats.
Note that this is not an exhaustive list. Processed profile format upgraders can be found in processed-profile-versioning.jt and gecko profile format upgraders can be found in gecko-profile-versioning.jt. Please refer to them for older upgraders or for exact implementations.
The SourceTable in profile.shared.sources was updated:
- The
uuidfield was renamed toid. - Two new fields
startLineandstartColumnwere added (1-based). These describe the start position of the script within its resource, useful for inline scripts. If the source covers the entire file, use(1, 1). - A new
sourceMapURLfield was added. Usenullfor entries with no source map URL.
The following tables have moved into profile.shared: stackTable, frameTable, funcTable, resourceTable, nativeSymbols. They are no longer per-thread.
Marker and sample data (including allocation samples) remains within the thread.
A new optional usedInnerWindowIDs field was added to the Thread type. This field contains an array of inner window IDs. It is used for the tab selector dropdown in the profiler UI, together with the information from profile.pages. When a tab is selected in this dropdown, threads that don't have an inner window ID for the selected tab in their usedInnerWindowIDs field are hidden. The array is treated as a set - the order of items in it has no meaning.
Profiles which don't use profile.pages also don't need to use the thread.usedInnerWindowIDs field.
A new SourceTable has been added to profile.shared.sources to centralize all source file information. The FuncTable.fileName field has been replaced with FuncTable.source, which references indices in the shared sources table. This change allows storing a UUID per JS source, which will be used for fetching sources.
The searchable property in marker schemas, originally added in version 44, is now removed again. Now all marker fields are searchable.
The stringArray is now shared across threads. The shared array is stored at profile.shared.stringArray.
Changes to the MarkerSchema type which is used for the elements of the array at profile.meta.markerSchema:
- A new
descriptionfield was added. This field is optional. - The
dataproperty was renamed tofields. - Every field must have a
keyand aformatproperty now. There are no static fields any more.
Concretely, this means that if you have a { "label": "Description", value: "..." } entry in your marker schema's data array, this entry needs to be removed and the description needs to be put into the description field instead, and the data property needs to be renamed to fields. If you have any other static fields, i.e. fields with label and value properties rather than key and format properties, then they need to be removed without replacement.
The implementation column was removed from the frameTable. Modern profiles from Firefox use subcategories to represent the information about the JIT type of a JS frame.
The optional meta.doesNotUseFrameImplementation field is no longer needed and was removed.
Furthermore, marker schema fields now support a hidden attribute. When present and set to true, such fields will be omitted from the tooltip and the sidebar.
And finally, profile.meta.sampleUnits.time now supports both 'ms' (milliseconds) and 'bytes'. When set to 'bytes', the time value of a sample will be interpreted as a bytes offset. This is useful for size profiles, where a sample's "time" describes the offset at which the piece is located within the entire file.
The columns category and subcategory were removed from the stackTable, to reduce the file size of profiles. The information in these columns was fully redundant with the category information in the frameTable. A stack's category and subcategory are determined as follows: If the stack's frame has a non-null category, then that's the stack's category, and the frame's subcategory (or 0 if null) becomes the stack's subcategory. Otherwise, if the stack is not a root node, it inherits the category and subcategory of its prefix stack. Otherwise, it defaults to the defaultCategory, which is defined as the first category in thread.meta.categories whose color is grey - at least one such category is required to be present. And the subcategory defaults to zero - all categories are required to have a "default" subcategory as their first subcategory.
The frameTable's category column now becomes essential. It was already required in the previous profile version, but the UI would mostly work even if it wasn't present. There are some existing profiles in rotation which are missing this column in the frameTable. The 52->53 upgrader fixes such profiles up by inferring it from the information in the stackTable.
No format changes, but a front-end behavior change: The schema for a marker is now looked up purely based on its data.type. In the past there were some special cases when data was null, or when data.type was tracing or Text. These special cases have been removed. The new behavior is simpler and more predictable, and was probably what you expected anyway.
This change came with a new version because we needed to upgrade old profiles from Firefox which were relying on the more complex behavior.
Two new marker schema field format types have been added: flow-id and terminating-flow-id, with string index values (like unique-string).
An optional isStackBased boolean field has been added to the marker schema.
The format can now optionally store sample and counter sample times as time deltas instead of absolute timestamps to reduce the JSON size.
A new sanitized-string marker schema format type has been added, allowing markers to carry arbitrary strings containing PII that will be sanitized along with URLs and FilePaths.
Removed the 'sampleGroups' object from the Counter structure.
The pid field of the Thread type is changed from string | number to string. The same happened to the data.otherPid field of IPC markers, and to the pid fields in the profiler.counters and profile.profilerOverhead lists.
An isMainThread field was added to the Thread type.
This replaces the following function:
export function isMainThread(thread: Thread): boolean {
return (
thread.name === 'GeckoMain' ||
// If the pid is a string, then it's not one that came from the system.
// These threads should all be treated as main threads.
typeof thread.pid === 'string' ||
// On Linux the tid of the main thread is the pid. This is useful for
// profiles imported from the Linux 'perf' tool.
String(thread.pid) === thread.tid
);
}The optimizations field is removed from the frameTable.
The searchable property is implemented in the marker schema. Previously all the name and category marker schema properties were automatically searchable and we had manual handling for FileIO, Log, DOMEvent, TraceEvent marker types.
The number property in counters is now optional.
The nativeSymbols table now has a new column: functionSize. Its values can be null.
The libs list has moved from Thread to Profile - it is now shared between all threads in the profile. And it only contains libs which are used by at least one resource.
The Lib fields have changed, too:
- The
start/end/offsetfields are gone: They are not needed after profile processing, when all frame addresses have been made library-relative; and these values usually differ in the different processes, so the fields could not have a meaningful value in the shared list. - There is a
codeIdfield which defaults tonull. This will be used in the future to store another ID which lets the symbol server look up correct binary. On Windows this is the dll / exe CodeId, and on Linux and Android this is the full ELF build ID.
We've also cleaned up the ResourceTable format:
- All resources now have names.
- Resources without a
hostorlibfield have these fields set tonullconsistently.
Older versions are not documented in this changelog but can be found in processed-profile-versioning.ts.
The SourceTable schema was updated:
- The
uuidfield was renamed toid. - Two new fields
startLineandstartColumnwere added (1-based). If the source covers the entire file, use(1, 1). - A new
sourceMapURLfield was added. It can be omitted if it is the last element in the row and there is no source map URL.
The sources field in the Gecko profile format is now non-optional. An upgrader was added that creates an empty SourceTable for profiles that don't have one.
frameTable location string field was changed to include an optional sourceIndex at the end of the string inside brackets for JS sources. For example, new JS frames look like this: functionName (http://script.url/:1234:1234)[1234] with the last number being its sourceIndex. This index references entries in the shared SourceTable in profile.sources (added in the same version) which centralizes all source file information.
Two new marker schema field format types have been added: flow-id and terminating-flow-id, with string index values (like unique-string).
An optional isStackBased boolean field has been added to the marker schema.
A new sanitized-string marker schema format type has been added, allowing markers to carry arbitrary strings containing PII that will be sanitized along with URLs and FilePaths.
Removed the 'sample_groups' object from the GeckoCounter structure.
A new unique-string marker schema format type has been added, allowing markers to carry unique-strings in their payloads.
The optimizations field is removed from the frameTable schema.
The searchable property is implemented in the marker schema. Previously all the name and category marker schema properties were automatically searchable and we had manual handling for FileIO, Log, DOMEvent, TraceEvent marker types.
Older versions are not documented in this changelog but can be found in gecko-profile-versioning.ts.